Anthony J. Clark
Last Updated: Sep 2024

Pomona College Computer Science 333 N. College Way Claremont, CA 91711
anthonyjclark.com Edmunds Hall 127

Education
Aug 2016
Ph.D. in Computer Science
Department of Computer Science and Engineering, Michigan State University
East Lansing, MI, USA
Dec 2009
B.S. in Computer Engineering
Department of Electrical and Computer Engineering, Kansas State University
Manhattan, KS, USA

Magna cum laude


Professional Experience
Jul 2020 to present
Assistant Professor of Computer Science
Department of Computer Science, Pomona College
Claremont, CA, USA

Research: mobile robotics • deep learning
Teaching: data structures • algorithms • neural networks • computer systems

Apr 2023 to Jun 2024
Visiting Associate
California Institute of Technology
Pasadena, CA, USA

Visiting researcher working in Dr. Soon-Jo Chung’s ARC Lab

Aug 2016 to May 2020
Assistant Professor of Computer Science
Department of Computer Science, Missouri State University
Springfield, MO, USA

Tenure-track assistant professor.

Jan to May 2016
Computer Science Teaching Assistant
Department of Computer Science and Engineering, Michigan State University
East Lansing, MI, USA

Administered two lab sections of Introduction to Programming II

May to Jul 2015
Computer Science Instructor
Department of Computer Science and Engineering, Michigan State University
East Lansing, MI, USA

Organized and taught Introduction to Programming II (CSE232) during the summer session.

May 2010 to Aug 2016
Graduate Fellow and Research Assistant
Department of Computer Science and Engineering, Michigan State University
East Lansing, MI, USA

Addressed optimization, adaptive control, and fabrication of bio-inspired mobile robotic systems.

May to Dec 2009
Undergraduate Research Assistant
Autonomous Vehicle Systems Laboratory, Kansas State University
Manhattan, KS, USA

Designed software used to capture images at specified GPS locations with an autonomous aerial vehicle.

Aug 2008 to May 2009
Undergraduate Research Assistant
Independent Research with Professor Stewart E. Stanton, Kansas State University
Manhattan, KS, USA

Investigated the fundamentals of convergence of complex solutions in power systems.

May to Aug 2008
Software Engineer, Intern
Department of Positioning and Sensors, Garmin International
Olathe, KS, USA

Solved problems associated with positioning error due to antenna performance.

May to Aug 2007
Undergraduate Research Fellow
Data Science Summer Institute, University of Illinois at Urbana-Champaign
Urbana-Champaign, IL, USA

Attended lectures covering the fundamentals of data science and worked on a team to create a reverse image search engine.

Aug 2007 to May 2009
SAS Tutor
Scholars Assisting Scholars (SAS) Program, Kansas State University
Manhattan, KS, USA

Attended lectures on the subject I was tutoring, provided walk-in, free tutoring consistent with course instruction, and led review sessions prior to exams.


Teaching and Course Development
Aug to Dec 2024
Mobile Robotics
  • Topics: design, fabrication, control, plannting, and ethics
  • Students per semester: 20
Aug 2022 to Dec 2024
Senior Seminar
Computer Science Department, Pomona College
Claremont, California, USA
  • Topics: Reading, discussion and presentation of research papers
  • Students per semester: 22, 10, 28
Jan 2021 to May 2023
Neural Networks
  • Topics: Neural networks foundations and ethics
  • Students per semester: 24, 36, 20, 37
Jan 2022 to Dec 2023
Computer Systems
Computer Science Department, Pomona College
Claremont, California, USA
  • Topics: Architecture, representations, concurrency, and I/O
  • Students per semester: 20, 26
Aug 2020 to Dec 2021
Algorithms
  • Topics: Asymptotic complexity, graphs, proofs, algorithm paradigms, computational complexity
  • Students per semester: 17, 24, 25
Aug to Dec 2020
Data Structures
  • Topics: Java, OOP, lists, big-o, trees
  • Students per semester: 26
Aug to Dec 2020
Independent Study
  • Topics: Deep learning, robotics, and graphics
  • Students per semester: 1
Jan to May 2020
CSC 125 Introduction to C++ Programming
Computer Science Department, Missouri State University
Springfield, Missouri, USA
  • Topics: C++, control flow, conditionals, memory management
  • Students per semester: 29
Aug 2016 to May 2020
CSC 325/611 Algorithms and Advanced Data Structures
  • Topics: Asymptotic complexity, graphs, proofs, algorithm paradigms, computational complexity
  • Students per semester: 27, 24, 23, 28, 29, 37, 41, 45
  • Student evaluation ratings (5-point scale): 4.75, 4.67, 4.89, 4.80, 4.89, 4.96, N/A
  • Prepare students for job interviews
  • This course includes 4 to 6 graduate students per semester
  • Developed all course materials
  • Created a syntax highlighter for pseudocode to maintain consistency on slides
  • Contribute to ABET accreditation through assessments
Aug 2016 to May 2020
CSC 333 Languages and Machines
  • Topics: formal languages, automata theory, programming languages, Unix
  • Students per semester: 25, 24, 21, 23, 21, 29, 30
  • Student evaluation ratings (5-point scale): 4.67, 4.77, 4.86, 4.73, 4.90, 4.87
  • Teach students about pair programming using Visual Studio Code Live Share
  • Developed all course materials
  • Contribute to ABET accreditation through assessments
May 2017 to May 2020
Independent Study
  • Mentored 24 independent study projects over the past eight semesters
  • Students range from seniors in the honors program to high school students attending Greenwood Elementary
  • Several research projects have been published in international computer science conferences
Aug 2017 to Dec 2019
CSC 742 Evolutionary Computing
  • Topics: genetic algorithms, evolutionary strategies, genetic programming, statistics
  • Students per semester: 14, 14
  • Student evaluation ratings (5-point scale): 4.39
  • Fall of even years
  • Developed all course materials
Jun to Aug 2019
CSC 232 Data Structures
  • Topics: C++ programming, array lists, linked lists, trees
  • Students per semester: 14
  • Student evaluation ratings (5-point scale): 4.89
  • Developed course materials with minimal assistance
Jan to May 2019
CSC 790 Deep Learning
  • Topics: convolutional neural networks, embeddings, optimization, regularization
  • Students per semester: 18
  • Student evaluation ratings (5-point scale): 4.85
  • Developed all course materials
Aug to Oct 2018
CSC 482 Seminar in Computer Science
  • Topics: interview preparation, ethics, teamwork
  • Students per semester: 35
Jan to May 2018
CSC 790 Advanced Robotics
  • Topics: robot operating system (ROS), computer vision, publisher-subscriber software
  • Students per semester: 14
  • Developed all course materials
May to Jul 2015
CSE 232, Introduction to Programming II
Department of Computer Science and Engineering, Michigan State University
East Lansing, MI, USA
  • Students per semester: 56
  • Student evaluation ratings (5-point scale): 4.61
  • Develop and present lectures covering introductory programming concepts using C++
  • Mentor and coordinate three teaching assistants

Awards, Honors, and Certificates
Oct 2018
Outstanding Reviewer
Elsevier
May 2018
Faculty Excellence in Teaching
College of Natural and Applied Sciences, Missouri State University

Nominated by the Computer Science Department and selected by the college awards committee.

Aug 2017
Master Advisor
Missouri State University

Completed the Advising Basics Workshop and the Master Advisor Workshop at Missouri State University. These workshops are day-long training sessions.

Jan 2017
Cultural Consciousness in the Classroom: Certificate of Participation
Missouri State University

Completed training for recruiting and retaining low-income students from historically underrepresented groups including first generation students.

May 2016
Outstanding Graduate Student Service Award
Department of Computer Science, Michigan State University
Sep 2013
Best Paper Award
Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems

Matthew J. Rose, Anthony J. Clark, Jared M. Moore, and Philip K. McKinley. Just Keep Swimming: Accounting for Uncertainty in Self-Modeling Aquatic Robots. In Proceedings of the 6th International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems, Taormina, Italy, September 2013.

Jul 2012
Best Paper Award
ALIFE Conference, Behavior and Intelligence Track

Anthony J. Clark, Jared Moore, Jianxun Wang, Xiaobo Tan, and Philip McKinley. Evolutionary design and experimental validation of a flexible caudal fin for robotic fish. In Proceedings of the Thirteenth International Conference on the Synthesis and Simulation of Living Systems, East Lansing, Michigan, USA, pages 325-332, July 2012.

Jan 2012
Honorable Mention: Graduate Research Fellowship Program
National Science Foundation
Aug 2010
Top Up Graduate Fellowship
NSF BEACON Center

Nominated by faculty at Michigan State University. This award was for $5,000 per year.

Aug 2010
University Enrichment Fellowship
Michigan State University

Nominated by the Computer Science Graduate Program at Michigan State University. This award guaranteed a research assistantship for six years.

Dec 2009
Graduated magna cum laude
Kansas State University
Aug 2008
Garmin Scholarship
Garmin International

Publications
Student authors are underlined.
Oct 2024

Semantics From Space: Satellite-Guided Thermal Semantic Segmentation Annotation for Aerial Field Robots

Connor Lee, Saraswati Soedarmadji, Matthew Anderson, Anthony J. Clark, and Soon-Jo Chung

IEEE/RSJ International Conference on Intelligent Robots and Systems. (IROS 2024), Abu Dhabi, UAE.

PDF Abstract Abstract: We present a new method to automatically generate semantic segmentation annotations for thermal imagery captured from an aerial vehicle by utilizing satellite-derived data products alongside onboard global positioning and attitude estimates. This new capability overcomes the challenge of developing thermal semantic perception algorithms for field robots due to the lack of annotated thermal field datasets and the time and costs of manual annotation, enabling precise and rapid annotation of thermal data from field collection efforts at a massively-parallelizable scale. By incorporating a thermal-conditioned refinement step with visual foundation models, our approach can produce highly-precise semantic segmentation labels using low-resolution satellite land cover data for little-to-no cost. It achieves 98.5 BibTeX @inproceedings{Lee.2024.IROS.SegmentationAnnotation, abstract = "We present a new method to automatically generate semantic segmentation annotations for thermal imagery captured from an aerial vehicle by utilizing satellite-derived data products alongside onboard global positioning and attitude estimates. This new capability overcomes the challenge of developing thermal semantic perception algorithms for field robots due to the lack of annotated thermal field datasets and the time and costs of manual annotation, enabling precise and rapid annotation of thermal data from field collection efforts at a massively-parallelizable scale. By incorporating a thermal-conditioned refinement step with visual foundation models, our approach can produce highly-precise semantic segmentation labels using low-resolution satellite land cover data for little-to-no cost. It achieves 98.5% of the performance from using costly high-resolution options and demonstrates between 70-160% improvement over popular zero-shot semantic segmentation methods based on large vision-language models currently used for generating annotations for RGB imagery. Code will be available at: https://github.com/connorlee77/aerial-auto-segment.", author = "Lee, Connor and Soedarmadji, Saraswati and Anderson, Matthew and Clark, Anthony J. and Chung, Soon-Jo", location = "Abu Dhabi, UAE", booktitle = "{IEEE/RSJ} International Conference on Intelligent Robots and Systems.", date = "2024-10-14", eventtitle = "{IROS} 2024", title = "Semantics from Space: Satellite-Guided Thermal Semantic Segmentation Annotation for Aerial Field Robots", }
Sep 2024

Training and Deploying Deep Learning Models for Real-Time Pathfinding in Indoor Environments

Aser Atawya, Kellie Au, Francisco Morales Puente, Tommy Ryan, Ella Zhu, and Anthony J. Clark

Southern California Robotics Symposium (SCR 2024), Riverside, California, USA.

PDF Abstract Abstract: Pathfinding in dynamic, indoor environments is fundamental to the reliable, safe, and real-time navigation of autonomous systems. In this study, we present our research generating datasets and comparing deep learning architectures for real-time pathfinding in indoor environments. We use simulation for data collection, compare six architectures, and analyze real-time inference performance on an NVIDIA JetBot. Our work showcases an end-to-end pathfinding approach and highlights challenges to address in future research. BibTeX @inproceedings{Clark.2024.SCRS.Pathfinding, abstract = "Pathfinding in dynamic, indoor environments is fundamental to the reliable, safe, and real-time navigation of autonomous systems. In this study, we present our research generating datasets and comparing deep learning architectures for real-time pathfinding in indoor environments. We use simulation for data collection, compare six architectures, and analyze real-time inference performance on an NVIDIA JetBot. Our work showcases an end-to-end pathfinding approach and highlights challenges to address in future research.", author = "Atawya, Aser and Au, Kellie and Morales Puente, Francisco and Ryan, Tommy and Zhu, Ella and Clark, Anthony J.", location = "Riverside, California, USA", booktitle = "Southern California Robotics Symposium", date = "2024-09-20", eventtitle = "{SCR} 2024", title = "Training and Deploying Deep Learning Models for Real-Time Pathfinding in Indoor Environments", }
Nov 2023

Development of a Hybrid Wheel/Leg Robot Using a Geared Coaxial Shaft Mechanism

James Clinton and Anthony J. Clark

Southern California Conference for Undergraduate Research (SCCUR 2023), Fullerton, California, USA.

PDF Abstract Abstract: Wheeled locomotion is a power-efficient, generally safer, and easy to control form of robot locomotion. Compared with legged or flying robots, however, wheeled robots struggle with uneven terrain and simple, medium-sized obstacles (i.e., obstacles roughly the size of their wheels). Legged locomotion solves this challenge by stepping or climbing over obstacles, but suffers from power consumption, poor stability, and control complexity. Hybrid wheel/leg robots transform their locomotion systems between wheeled and legged modes to maximize the performance benefits of each locomotion mode. We present our design of a novel hybrid wheel/leg robot that uses a geared coaxial shaft mechanism to transform between states of wheeled and legged-wheel locomotion. The mechanism is also able to operate at intermediate states between these two modes. We present our mechanical design, development of a basic control policy, and preliminary experiments demonstrating basic capabilities of the robot. We show that the robot is able to navigate flat terrain in the wheeled locomotion mode, and can make use of hybrid legged locomotion modes to navigate rough terrain and obstacles. In the future, we plan to iterate on the mechanism’s design, develop and evaluate more advanced control policies that leverage the mechanism’s intermediate locomotion modes, and evaluate the mechanism’s performance in real-world environments. BibTeX @inproceedings{Clinton.2023.SCCUR.HybridWheelLeg, abstract = " Wheeled locomotion is a power-efficient, generally safer, and easy to control form of robot locomotion. Compared with legged or flying robots, however, wheeled robots struggle with uneven terrain and simple, medium-sized obstacles (i.e., obstacles roughly the size of their wheels). Legged locomotion solves this challenge by stepping or climbing over obstacles, but suffers from power consumption, poor stability, and control complexity. Hybrid wheel/leg robots transform their locomotion systems between wheeled and legged modes to maximize the performance benefits of each locomotion mode. We present our design of a novel hybrid wheel/leg robot that uses a geared coaxial shaft mechanism to transform between states of wheeled and legged-wheel locomotion. The mechanism is also able to operate at intermediate states between these two modes. We present our mechanical design, development of a basic control policy, and preliminary experiments demonstrating basic capabilities of the robot. We show that the robot is able to navigate flat terrain in the wheeled locomotion mode, and can make use of hybrid legged locomotion modes to navigate rough terrain and obstacles. In the future, we plan to iterate on the mechanism's design, develop and evaluate more advanced control policies that leverage the mechanism's intermediate locomotion modes, and evaluate the mechanism's performance in real-world environments.", author = "Clinton, James and Clark, Anthony J.", location = "Fullerton, California, USA", booktitle = "Southern California Conference for Undergraduate Research", date = "2023-11-18", eventtitle = "{SCCUR} 2023", title = "Development of a Hybrid Wheel/Leg Robot Using a Geared Coaxial Shaft Mechanism", }
Sep 2023

Creating Dynamic Simulation Environments With Unreal Engine 5

Daisy Abbott, Anjali Nuggehalli, Francisco Morales Puente, Chau Vu, Ella Zhu, and Anthony J. Clark

Southern California Robotics Symposium (SCR 2023), Irvine, California, USA.

PDF Abstract Abstract: Simulation is a vital component for many machine learning-based systems. In this abstract, we present our work using Unreal Engine 5 to create realistic environments for training a neural network used in the navigation system of a mobile robot. We explore the use of randomized textures to create dynamic environments, and we evaluate trained models in environments with both randomly changing and static textures. BibTeX @inproceedings{MoralesPuente.2023.SCRS.UE5Simulation, abstract = "Simulation is a vital component for many machine learning-based systems. In this abstract, we present our work using Unreal Engine 5 to create realistic environments for training a neural network used in the navigation system of a mobile robot. We explore the use of randomized textures to create dynamic environments, and we evaluate trained models in environments with both randomly changing and static textures.", author = "Abbott, Daisy and Nuggehalli, Anjali and Morales Puente, Francisco and Vu, Chau and Zhu, Ella and Clark, Anthony J.", location = "Irvine, California, USA", booktitle = "Southern California Robotics Symposium", date = "2023-09-22", eventtitle = "{SCR} 2023", title = "Creating Dynamic Simulation Environments With Unreal Engine 5", }
Jul 2023

Does Kinematic-Based Pretraining Improve Evolution of Quadrupedal Gaits?

Kevin J. Ayala Ahumada, Jared M. Moore, and Anthony J. Clark

Conference on Artificial Life (ALIFE 2023), Sapporo, Japan.

PDF Abstract Abstract: Neural networks are often chosen as controllers in evolutionary robotics. In all but a few cases, neural networks are evolved from scratch. In this study, we investigate the effect of pretraining neural networks using a biologically inspired walking gait. We first generate joint angles for a walking gait using an inverse kinematics model. We then train a conventional feed-forward neural network to reproduce these joint angles. The pretrained model is used to seed an initial population of neural networks, which are coevolved along with the morphology of a quadrupedal robot using Lexicase selection. Our initial results show that while pretraining does not necessarily lead to higher fitness at the end of evolution, it does lead to more consistent performance and more lifelike final behaviors. This exploration has left us with many questions about the importance and process of pretraining in evolutionary robotics, and we believe our results suggest the technique is worth further investigation. BibTeX @inproceedings{Ayala.2023.ALIFE.Pretraining, abstract = "Neural networks are often chosen as controllers in evolutionary robotics. In all but a few cases, neural networks are evolved from scratch. In this study, we investigate the effect of pretraining neural networks using a biologically inspired walking gait. We first generate joint angles for a walking gait using an inverse kinematics model. We then train a conventional feed-forward neural network to reproduce these joint angles. The pretrained model is used to seed an initial population of neural networks, which are coevolved along with the morphology of a quadrupedal robot using Lexicase selection. Our initial results show that while pretraining does not necessarily lead to higher fitness at the end of evolution, it does lead to more consistent performance and more lifelike final behaviors. This exploration has left us with many questions about the importance and process of pretraining in evolutionary robotics, and we believe our results suggest the technique is worth further investigation.", author = "Ayala Ahumada, Kevin J. and Moore, Jared M. and Clark, Anthony J.", location = "Sapporo, Japan", publisher = "MIT Press", booktitle = "Conference on Artificial Life", date = "2023-07-24", doi = "", eventtitle = "{ALIFE} 2023", title = "Does Kinematic-Based Pretraining Improve Evolution of Quadrupedal Gaits?", }
Sep 2022

Searching for Problematic Simulation Conditions

Elizabeth Johnson, Simon Heck, Keneth Gonzalez Hernandez, and Anthony J. Clark

Southern California Robotics Symposium (SCR 2022), Los Angeles, California, USA.

PDF Abstract Abstract: Many robot use-cases put a robot in close contact with people. These scenarios require the robot to make complex decisions that—above all else—must be safe. Often, sensor processing and decision making methods rely heavily on machine learning, and these techniques are only as useful as the training dataset. Current methods and datasets do not account for enough variation or extraordinary conditions. We propose using novelty search to discover scenarios causing a model to behave poorly. BibTeX @inproceedings{Johnson.2022.SCRS.Problematic, abstract = "Many robot use-cases put a robot in close contact with people. These scenarios require the robot to make complex decisions that---above all else---must be safe. Often, sensor processing and decision making methods rely heavily on machine learning, and these techniques are only as useful as the training dataset. Current methods and datasets do not account for enough variation or extraordinary conditions. We propose using novelty search to discover scenarios causing a model to behave poorly.", author = "Johnson, Elizabeth and Heck, Simon and Hernandez, Keneth Gonzalez and Clark, Anthony J.", location = "Los Angeles, California, USA", booktitle = "Southern California Robotics Symposium", date = "2022-09-22", eventtitle = "{SCR} 2022", title = "Searching for Problematic Simulation Conditions", }
Dec 2021

Investigating Neural Network Architectures, Techniques, and Datasets for Autonomous Navigation in Simulation

Oliver Chang, Christiana Marchese, Jared Mejia, and Anthony J. Clark

IEEE Symposium Series on Computational Intelligence (SSCI 2021), Orlando, Florida, USA.

PDF Abstract Abstract: Neural networks (NNs) are becoming an increas- ingly important part of mobile robot control systems. Com- pared with traditional methods, NNs (and other data-driven techniques) produce comparable—if not better—results while requiring less engineering knowhow. Training NNs, however, still requires exploration of a significant number of architectural, optimization, and evaluation options. In this study, we build a simulation environment, generate three image datasets using distinct techniques, train 652 models (including replicates) using a variety of architectures and paradigms (e.g., classification, regression, etc.), and evaluate the navigation ability of the model in simulation. Our goal is to explore a large number of model possibilities so that we can select the most promising for future study with a physical device. Training datasets leading to the best performing models were those that included a significant amount of noise from seemingly inefficient actions. The most promising models explicitly incorporated “memory” wherein previous actions were included as an input in the next step. Such models performed as good or better than conventional convolutional NNs, recurrent NNs, and custom architectures including two camera frames. Although trained models perform well in an environment matching the distribution of the training dataset, they fail when the simulation environment is altered in a seemingly insignificant manner. In robotics research it is often taken for granted that a model with good validation characteristics will perform well on the underlying task, but the results presented here show that there can often be a loose relationship between validation metrics and performance. BibTeX @inproceedings{Chang.2021.SSCI.Architectures, abstract = "Neural networks (NNs) are becoming an increas- ingly important part of mobile robot control systems. Com- pared with traditional methods, NNs (and other data-driven techniques) produce comparable—if not better—results while requiring less engineering knowhow. Training NNs, however, still requires exploration of a significant number of architectural, optimization, and evaluation options. In this study, we build a simulation environment, generate three image datasets using distinct techniques, train 652 models (including replicates) using a variety of architectures and paradigms (e.g., classification, regression, etc.), and evaluate the navigation ability of the model in simulation. Our goal is to explore a large number of model possibilities so that we can select the most promising for future study with a physical device. Training datasets leading to the best performing models were those that included a significant amount of noise from seemingly inefficient actions. The most promising models explicitly incorporated “memory” wherein previous actions were included as an input in the next step. Such models performed as good or better than conventional convolutional NNs, recurrent NNs, and custom architectures including two camera frames. Although trained models perform well in an environment matching the distribution of the training dataset, they fail when the simulation environment is altered in a seemingly insignificant manner. In robotics research it is often taken for granted that a model with good validation characteristics will perform well on the underlying task, but the results presented here show that there can often be a loose relationship between validation metrics and performance.", author = "Chang, Oliver and Marchese, Christiana and Mejia, Jared and Clark, Anthony J.", location = "Orlando, Florida, USA", publisher = "{IEEE}", booktitle = "{IEEE} Symposium Series on Computational Intelligence", date = "2021-12-01", doi = "", eventtitle = "{SSCI} 2021", isbn = "", title = "Investigating Neural Network Architectures, Techniques, and Datasets for Autonomous Navigation in Simulation", }
Jul 2021

Supervision and Evolution: Pretraining Neural Networks for Quadrupedal Locomotion

Jared M. Moore and Anthony J. Clark

Conference on Artificial Life (ALIFE 2021), Online. DOI: 10.1162/isal_a_00363

PDF DOI Abstract Abstract: Neural networks (NNs) are effective controllers for evolutionary robotics, imposing few limits on potential gaits. Morphology evolved with a controller enables brain and body to become tightly coupled. Typically, NN parameters (sometimes architectures) and animat bodies are randomly initialized at the start of evolution. In this paper, we pretrain NNs with supervised learning, bootstrapping NN outputs towards oscillating behaviors prior to evolution. We focus on quadrupedal gaits as they are well-studied in biology and several common gait patterns have been identified, named, and studied by the research community. We hypothesize that performance of evolved gaits will improve with pretraining compared to beginning evolution with randomly initialized NNs. Our results show that only some pretraining regimens outperform (in terms of distance traveled and viability) random initialization of NN parameters. Furthermore, some regimens introduce an initial bias that is difficult to overcome, resulting in better initial performance but worse performance in the long term. BibTeX @inproceedings{Moore.2021.ALIFE.Pretrain, abstract = "Neural networks (NNs) are effective controllers for evolutionary robotics, imposing few limits on potential gaits. Morphology evolved with a controller enables brain and body to become tightly coupled. Typically, NN parameters (sometimes architectures) and animat bodies are randomly initialized at the start of evolution. In this paper, we pretrain NNs with supervised learning, bootstrapping NN outputs towards oscillating behaviors prior to evolution. We focus on quadrupedal gaits as they are well-studied in biology and several common gait patterns have been identified, named, and studied by the research community. We hypothesize that performance of evolved gaits will improve with pretraining compared to beginning evolution with randomly initialized NNs. Our results show that only some pretraining regimens outperform (in terms of distance traveled and viability) random initialization of NN parameters. Furthermore, some regimens introduce an initial bias that is difficult to overcome, resulting in better initial performance but worse performance in the long term.", author = "Moore, Jared M. and Clark, Anthony J.", location = "Online", publisher = "MIT Press", booktitle = "Conference on Artificial Life", date = "2021-07-19", doi = "10.1162/isal_a_00363", eventtitle = "{ALIFE} 2021", title = "Supervision and Evolution: Pretraining Neural Networks for Quadrupedal Locomotion", }
Jul 2020

MorphWorld: A State Transition Simulator

Matthew Shan, Jared M. Moore, and Anthony J. Clark

Conference on Artificial Life (ALIFE 2020), Montreal, CA (Remote Conference). DOI: 10.1162/isal_a_00253

PDF DOI Abstract Abstract: Digital simulation enables a wide variety of research and applications underlying the study of artificial life. In evolutionary robotics applications, the focus is often on maximizing performance of an animat for a specific task. Analyzing evolved behaviors can be challenging, however, given the complex coupling of morphology and brain. In this paper, we introduce a simulation environment built to investigate animats capable of smoothly transitioning between operating modes (e.g., from cautious to aggressive or from one physical form to another). The simulator provides functionality for logging sensory information as well as animat state enabling a deep analysis. Although more abstract than soft-body or rigid-body physics engines, it is lightweight and efficient, allowing for a high number of simulations in a small amount of time. The simulation supplements other more complex physics-based environments providing for greater inspection of sensor information and animat behavior. Furthermore, it is designed to provide an extensible test bed beyond just gait transitions to assess new artificial intelligence and evolutionary algorithms and more importantly the combination of these techniques. BibTeX @inproceedings{Shan.2020.ALIFE.MorphWorld, abstract = "Digital simulation enables a wide variety of research and applications underlying the study of artificial life. In evolutionary robotics applications, the focus is often on maximizing performance of an animat for a specific task. Analyzing evolved behaviors can be challenging, however, given the complex coupling of morphology and brain. In this paper, we introduce a simulation environment built to investigate animats capable of smoothly transitioning between operating modes (e.g., from cautious to aggressive or from one physical form to another). The simulator provides functionality for logging sensory information as well as animat state enabling a deep analysis. Although more abstract than soft-body or rigid-body physics engines, it is lightweight and efficient, allowing for a high number of simulations in a small amount of time. The simulation supplements other more complex physics-based environments providing for greater inspection of sensor information and animat behavior. Furthermore, it is designed to provide an extensible test bed beyond just gait transitions to assess new artificial intelligence and evolutionary algorithms and more importantly the combination of these techniques.", author = "Shan, Matthew and Moore, Jared M. and Clark, Anthony J.", location = "Montreal, CA (Remote Conference)", publisher = "MIT Press", booktitle = "Conference on Artificial Life", date = "2020-07-13", doi = "10.1162/isal_a_00253", eventtitle = "{ALIFE} 2020", pages = "747--749", title = "MorphWorld: A State Transition Simulator", }
Sep 2019

Comparing CNN Inputs for Terrain Classification Using Simulation

Anthony J. Clark, Jesse Simpson, and Jared Hall

IEEE Transdisciplinary AI (TransAI 2019), Laguna Hills, California, USA. DOI: 10.1109/TransAI46475.2019.00015

PDF DOI Abstract Abstract: Mobile robots frequently operate in rough, uneven terrain. One way for them to identify easier to traverse paths is to use deep learning methods, such as a convolutional neural network (CNN). It is not clear, however, what input should be provided to the CNN to best enable it to classify different types of terrain. In this study, we investigate and compare several different inputs formats for improving terrain classification using a CNN. All experiments take place in simulation, where we have complete control over terrain (e.g., shapes and textures) and information about our robot. Our experiments lead us to the following: (1) input formats should prefer grayscale over color images as color has a tendency to overfit the training data, and (2) disparity maps also improve classification compared with raw image data. These results can be used to improve the performance of terrain classification; particularly as they apply to transformable-wheel robots. BibTeX @inproceedings{Clark.2019.TA.ComparingCNNInputs, abstract = "Mobile robots frequently operate in rough, uneven terrain. One way for them to identify easier to traverse paths is to use deep learning methods, such as a convolutional neural network (CNN). It is not clear, however, what input should be provided to the CNN to best enable it to classify different types of terrain. In this study, we investigate and compare several different inputs formats for improving terrain classification using a CNN. All experiments take place in simulation, where we have complete control over terrain (e.g., shapes and textures) and information about our robot. Our experiments lead us to the following: (1) input formats should prefer grayscale over color images as color has a tendency to overfit the training data, and (2) disparity maps also improve classification compared with raw image data. These results can be used to improve the performance of terrain classification; particularly as they apply to transformable-wheel robots.", author = "Clark, Anthony J. and Simpson, Jesse and Hall, Jared", location = "Laguna Hills, California, USA", booktitle = "{IEEE} Transdisciplinary {AI}", date = "2019-09-25", doi = "10.1109/TransAI46475.2019.00015", eventtitle = "{TransAI} 2019", isbn = "978-1-7281-4127-5", pages = "43--47", title = "Comparing {CNN} Inputs for Terrain Classification Using Simulation", }
Sep 2019

Construct of Sarcasm on Social Media Platform

Dipto Das and Anthony J. Clark

IEEE International Conference on Humanized Computing and Communication (HCC 2019), Laguna Hills, California, USA. DOI: 10.1109/HCC46620.2019.00023

PDF DOI Slides Abstract Abstract: The basic idea behind machine learning-based systems, or artificial intelligence in general, is mimicking how humans operate. This idea is particularly true for our problem, sarcasm detection on social networking sites (SNSs). Therefore, before proceeding to build a system that can detect sarcasm on SNSs, we attempt to understand how humans do the same. Many studies propose approaches based on personal experience and word-level definition of "sarcasm". However, in this paper, we aim to find more general themes that are typical with users while detecting and expressing sarcasm on SNSs through a qualitative study to build a more effective sarcasm detection model. BibTeX @inproceedings{Das.2019.HCC.ConstructSarcasmSocial, abstract = "The basic idea behind machine learning-based systems, or artificial intelligence in general, is mimicking how humans operate. This idea is particularly true for our problem, sarcasm detection on social networking sites (SNSs). Therefore, before proceeding to build a system that can detect sarcasm on SNSs, we attempt to understand how humans do the same. Many studies propose approaches based on personal experience and word-level definition of {"}sarcasm{"}. However, in this paper, we aim to find more general themes that are typical with users while detecting and expressing sarcasm on SNSs through a qualitative study to build a more effective sarcasm detection model.", author = "Das, Dipto and Clark, Anthony J.", location = "Laguna Hills, California, USA", booktitle = "{IEEE} International Conference on Humanized Computing and Communication", date = "2019-09-25", doi = "10.1109/HCC46620.2019.00023", eventtitle = "{HCC} 2019", isbn = "978-1-7281-4125-1", pages = "106--113", title = "Construct of Sarcasm on Social Media Platform", }
Sep 2019

Satire vs Fake News: You Can Tell by the Way They Say It

Dipto Das and Anthony J. Clark

IEEE Transdisciplinary AI (TransAI 2019), Laguna Hills, California, USA. DOI: 10.1109/TransAI46475.2019.00012

PDF DOI Slides Abstract Abstract: In recent times, "fake news" has become an increasingly important concept. Primarily, because information is now able to more quickly and deeply propagate among users due to the pervasive nature of the Internet and digital media. That is why it has recently received a large amount of attention from computer science researchers. A large number of studies demonstrate different methods for detecting misinformation in contents shared on the Internet. On the other hand, satire and irony as a part of usual human communication have received less attention. Whereas fake news means misinformation meant to deceive people, satire is misinformation meant to entertain or criticize. Thus, despite both satire and fake news being misinformation these two concepts have different objectives and impacts. Currently, only a few studies have focused on differentiating between satire and fake news. In this paper, we present the limitations of existing works for classifying satire and fake news; discuss the feasibility of using a subjective concept like storytelling as a way to classify satire and fake news; and present a supervised learning approach to classify satire and fake news. BibTeX @inproceedings{Das.2019.TA.SatireVsFake, abstract = "In recent times, {"}fake news{"} has become an increasingly important concept. Primarily, because information is now able to more quickly and deeply propagate among users due to the pervasive nature of the Internet and digital media. That is why it has recently received a large amount of attention from computer science researchers. A large number of studies demonstrate different methods for detecting misinformation in contents shared on the Internet. On the other hand, satire and irony as a part of usual human communication have received less attention. Whereas fake news means misinformation meant to deceive people, satire is misinformation meant to entertain or criticize. Thus, despite both satire and fake news being misinformation these two concepts have different objectives and impacts. Currently, only a few studies have focused on differentiating between satire and fake news. In this paper, we present the limitations of existing works for classifying satire and fake news; discuss the feasibility of using a subjective concept like storytelling as a way to classify satire and fake news; and present a supervised learning approach to classify satire and fake news.", author = "Das, Dipto and Clark, Anthony J.", location = "Laguna Hills, California, USA", booktitle = "{IEEE} Transdisciplinary {AI}", date = "2019-09-25", doi = "10.1109/TransAI46475.2019.00012", eventtitle = "{TransAI} 2019", isbn = "978-1-7281-4127-5", pages = "22--26", title = "Satire vs Fake News: You Can Tell by the Way They Say It", }
Sep 2019

Understanding the Attention Model of Humans in Sarcastic Videos

Dipto Das, Md Forhad Hossain, and Anthony J. Clark

IEEE Transdisciplinary AI (TransAI 2019), Laguna Hills, California, USA. DOI: 10.1109/TransAI46475.2019.00022

PDF DOI Abstract Abstract: Sarcasm is a usual part of human communication that has long been ignored by sentiment analysis researchers. Sarcasm is also an important aspect in entertainment industry for TV series, movies etc. to gain popularity. Very recently, some works have showed the applicability of multimodality (e.g., image, text) in sarcasm research from a sentiment analysis perspective instead of only text-based approaches. However, none of those harnesses video data. We argue videos can be interesting to study to understand nature of sarcasm on social media. We are interested to study how sarcastic videos gain individual’s attention and popularity at large. As an application of this, we showed how an AI agent can suggest about possible areas to gain viewers’ attention in a directed sarcastic video. Identification of both attention gaining areas (AGA) and objects contained in sarcastic videos can be compared with the AGAs and objects in previously successful/popular sarcastic videos. Such AI agent can help inexperienced directors in entertainment industry as a guide and experienced ones to study the changes brought over time in this regard. In this paper, we present two AI agents to identify the optimal AGAs and one empirical study of objects commonly shown in directed sarcastic video settings. BibTeX @inproceedings{Das.2019.TA.UnderstandingAttentionModel, abstract = "Sarcasm is a usual part of human communication that has long been ignored by sentiment analysis researchers. Sarcasm is also an important aspect in entertainment industry for TV series, movies etc. to gain popularity. Very recently, some works have showed the applicability of multimodality (e.g., image, text) in sarcasm research from a sentiment analysis perspective instead of only text-based approaches. However, none of those harnesses video data. We argue videos can be interesting to study to understand nature of sarcasm on social media. We are interested to study how sarcastic videos gain individual's attention and popularity at large. As an application of this, we showed how an AI agent can suggest about possible areas to gain viewers' attention in a directed sarcastic video. Identification of both attention gaining areas (AGA) and objects contained in sarcastic videos can be compared with the AGAs and objects in previously successful/popular sarcastic videos. Such AI agent can help inexperienced directors in entertainment industry as a guide and experienced ones to study the changes brought over time in this regard. In this paper, we present two AI agents to identify the optimal AGAs and one empirical study of objects commonly shown in directed sarcastic video settings.", author = "Das, Dipto and Hossain, Md Forhad and Clark, Anthony J.", location = "Laguna Hills, California, USA", booktitle = "{IEEE} Transdisciplinary {AI}", date = "2019-09-25", doi = "10.1109/TransAI46475.2019.00022", eventtitle = "{TransAI} 2019", isbn = "978-1-7281-4127-5", pages = "84--87", title = "Understanding the Attention Model of Humans in Sarcastic Videos", }
Jul 2019

Improve Quadrupedal Locomotion With Actuated or Passive Joints?

Jared M. Moore and Anthony J. Clark

Conference on Artificial Life (ALIFE 2019), Newcastle, United Kingdom. DOI: 10.1162/isal_a_00221

PDF DOI Abstract Abstract: Animals interact with their environment softly through interaction of muscles, tendons, and rigid skeleton. By incorporating flexibility, they reduce ground impact forces and improve locomotive efficiency. Flexibility is also beneficial for robotic systems, although it remains challenging to implement. In this paper, we explore the addition of passive flexibility to a quadrupedal animat; we measure the impact of flexibility on both locomotive performance and energy efficiency of movement. Results show that spine and lower limb flexibility can significantly increase distance traveled when compared to an animat with no flexibility. However, replacing passively flexibile joints with actively controlled joints evolves more effective individuals with similar efficiency. Given these results, the number of joints and joint configuration appear to drive performance increases rather than just the addition of passive flexibility. BibTeX @inproceedings{Moore.2019.ALIFE.ImproveQuadrupedalLocomotion, abstract = "Animals interact with their environment softly through interaction of muscles, tendons, and rigid skeleton. By incorporating flexibility, they reduce ground impact forces and improve locomotive efficiency. Flexibility is also beneficial for robotic systems, although it remains challenging to implement. In this paper, we explore the addition of passive flexibility to a quadrupedal animat; we measure the impact of flexibility on both locomotive performance and energy efficiency of movement. Results show that spine and lower limb flexibility can significantly increase distance traveled when compared to an animat with no flexibility. However, replacing passively flexibile joints with actively controlled joints evolves more effective individuals with similar efficiency. Given these results, the number of joints and joint configuration appear to drive performance increases rather than just the addition of passive flexibility.", author = "Moore, Jared M. and Clark, Anthony J.", location = "Newcastle, United Kingdom", publisher = "MIT Press", booktitle = "Conference on Artificial Life", date = "2019-07-29", doi = "10.1162/isal_a_00221", eventtitle = "{ALIFE} 2019", isbn = "978-0-262-35844-6", pages = "559--566", title = "Improve Quadrupedal Locomotion with Actuated or Passive Joints?", }
Dec 2018

Evolving Controllers for a Transformable Wheel Mobile Robot

Anthony J. Clark, Keith A. Cissell, and Jared M. Moore

Complexity. DOI: 10.1155/2018/7692042

PDF DOI Abstract Abstract: Unmanned ground vehicles (UGVs) are well suited to tasks that are either too dangerous or too monotonous for people. For example, UGVs can traverse arduous terrain in search of disaster victims. However, it is difficult to design these systems so that they perform well in a variety of different environments. In this study, we evolve controllers and physical characteristics of a UGV with transformable wheels to improve its mobility in a simulated environment. The UGV’s mission is to visit a sequence of coordinates while automatically handling obstacles of varying sizes by extending wheel struts radially outward from the center of each wheel. Evolved finite state machines (FSMs) and artificial neural networks (ANNs) are compared, and a set of controller design principles are gathered from analyzing these experiments. Results show similar performance between FSM and ANN controllers but differing strategies. Finally, we show that a UGV’s controller and physical characteristics can be effectively chosen by examining results from evolutionary optimization. BibTeX @article{Clark.2018.Complexity.EvolvingControllersTransformable, abstract = "Unmanned ground vehicles (UGVs) are well suited to tasks that are either too dangerous or too monotonous for people. For example, UGVs can traverse arduous terrain in search of disaster victims. However, it is difficult to design these systems so that they perform well in a variety of different environments. In this study, we evolve controllers and physical characteristics of a UGV with transformable wheels to improve its mobility in a simulated environment. The UGV’s mission is to visit a sequence of coordinates while automatically handling obstacles of varying sizes by extending wheel struts radially outward from the center of each wheel. Evolved finite state machines (FSMs) and artificial neural networks (ANNs) are compared, and a set of controller design principles are gathered from analyzing these experiments. Results show similar performance between FSM and ANN controllers but differing strategies. Finally, we show that a UGV’s controller and physical characteristics can be effectively chosen by examining results from evolutionary optimization.", author = "Clark, Anthony J. and Cissell, Keith A. and Moore, Jared M.", date = "2018-12-16", doi = "10.1155/2018/7692042", issn = "1076-2787, 1099-0526", journaltitle = "Complexity", langid = "english", title = "Evolving Controllers for a Transformable Wheel Mobile Robot", volume = "2018", }
Dec 2018

An Ensemble of Face Recognition Algorithms for Unsupervised Expansion of Training Data

Jeffrey Dale and Anthony J. Clark

International Conference on Computational Science and Computational Intelligence (CSCI 2018), Las Vegas, Nevada, USA. DOI: 10.1109/CSCI.2018.00072

PDF DOI Slides Abstract Abstract: Facial recognition is a classical problem in computer vision. The accuracy of face recognition algorithms is crucial in practice, as systems are increasingly secured with biometric locks. However, the performance of these algorithms is heavily dependent upon the size of the training data. This paper proposes an unsupervised ensemble method for expanding the set of training faces when only a single labeled face per subject is known. We show that the ensemble’s confidence measure is sufficient to expand the training set to the point where more sophisticated algorithms can take over in classification. BibTeX @inproceedings{Dale.2018.CSCI.EnsembleFaceRecognition, abstract = "Facial recognition is a classical problem in computer vision. The accuracy of face recognition algorithms is crucial in practice, as systems are increasingly secured with biometric locks. However, the performance of these algorithms is heavily dependent upon the size of the training data. This paper proposes an unsupervised ensemble method for expanding the set of training faces when only a single labeled face per subject is known. We show that the ensemble’s confidence measure is sufficient to expand the training set to the point where more sophisticated algorithms can take over in classification.", author = "Dale, Jeffrey and Clark, Anthony J.", location = "Las Vegas, Nevada, USA", booktitle = "International Conference on Computational Science and Computational Intelligence", date = "2018-12-13", doi = "10.1109/CSCI.2018.00072", eventtitle = "{CSCI} 2018", isbn = "978-1-72811-360-9", langid = "english", title = "An Ensemble of Face Recognition Algorithms for Unsupervised Expansion of Training Data", }
Oct 2018

Sarcasm Detection on Facebook: A Supervised Learning Approach

Dipto Das and Anthony J. Clark

International Conference on Multimodal Interaction Adjunct (ICMI 2018), Boulder, Colorado, USA. DOI: 10.1145/3281151.3281154

PDF DOI Abstract Abstract: Sarcasm is a common feature of user interaction on social networking sites. Sarcasm differs with typical communication in alignment of literal meaning with intended meaning. Humans can recognize sarcasm from sufficient context information including from the various contents available on SNS. Existing literature mainly uses text data to detect sarcasm; though, a few recent studies propose to use image data. To date, no study has focused on user interaction pattern as a source of context information for detecting sarcasm. In this paper, we present a supervised machine learning based approach focusing on both contents of posts (e.g., text, image) and users’ interaction on those posts on Facebook. BibTeX @inproceedings{Das.2018.ICMIA.SarcasmDetectionFacebook, abstract = "Sarcasm is a common feature of user interaction on social networking sites. Sarcasm differs with typical communication in alignment of literal meaning with intended meaning. Humans can recognize sarcasm from sufficient context information including from the various contents available on SNS. Existing literature mainly uses text data to detect sarcasm; though, a few recent studies propose to use image data. To date, no study has focused on user interaction pattern as a source of context information for detecting sarcasm. In this paper, we present a supervised machine learning based approach focusing on both contents of posts (e.g., text, image) and users' interaction on those posts on Facebook.", author = "Das, Dipto and Clark, Anthony J.", location = "Boulder, Colorado, USA", publisher = "ACM Press", booktitle = "International Conference on Multimodal Interaction Adjunct", date = "2018-10-16", doi = "10.1145/3281151.3281154", eventtitle = "{ICMI} 2018", isbn = "978-1-4503-6002-9", shorttitle = "Sarcasm Detection on {Facebook}", title = "Sarcasm Detection on {Facebook}: A Supervised Learning Approach", }
Sep 2018

Sarcasm Detection on Flickr Using a CNN

Dipto Das and Anthony J. Clark

International Conference on Computing and Big Data (ICCBD 2018), Charleston, South Carolina, USA. DOI: 10.1145/3277104.3277118

PDF DOI Abstract Abstract: Sarcasm is an important aspect of human communication. However, it is often difficult to detect or understand this sentiment because the literal meaning conveyed in communication is opposite of the intended meaning. Though the field of sentiment analysis is well studied, sarcasm has often been ignored by the research community. So far, to detect sarcasm on social media, studies have largely focused upon textual features. However, visual cues are an important part of sarcasm. In this paper, we present a convolutional neural network based model for detecting sarcasm based on images shared on a popular social photo sharing site, Flickr. BibTeX @inproceedings{Das.2018.ICCBD.SarcasmDetectionFlickr, abstract = "Sarcasm is an important aspect of human communication. However, it is often difficult to detect or understand this sentiment because the literal meaning conveyed in communication is opposite of the intended meaning. Though the field of sentiment analysis is well studied, sarcasm has often been ignored by the research community. So far, to detect sarcasm on social media, studies have largely focused upon textual features. However, visual cues are an important part of sarcasm. In this paper, we present a convolutional neural network based model for detecting sarcasm based on images shared on a popular social photo sharing site, Flickr.", author = "Das, Dipto and Clark, Anthony J.", location = "Charleston, South Carolina, USA", publisher = "ACM Press", booktitle = "International Conference on Computing and Big Data", date = "2018-09-08", doi = "10.1145/3277104.3277118", eventtitle = "{ICCBD} 2018", isbn = "978-1-4503-6540-6", pages = "56--61", title = "Sarcasm Detection on {F}lickr Using a {CNN}", }
Jul 2018

Review: A Web-Based Simulation Viewer for Sharing Evolutionary Robotics Results

Anthony J. Clark and Jared M. Moore

Genetic and Evolutionary Computation Conference (GECCO 2018), Kyoto, Japan. DOI: 10.1145/3205651.3208292

PDF DOI Abstract Abstract: Evolutionary robotics researchers often need to share results that may be too difficult to describe in text and too complex to show using images. Many researchers include links to videos as supplementary materials, but videos have a predefined view of the scene and do not allow watchers to adjust the viewing angle to their preference. In this paper we present a web-based application (based on three.js) for sharing interactive animations. Specifically, our tool (called Review) enables researchers to generate simple animation log data that can be loaded in any modern web browser on a computer or mobile device. The camera in these animations is controlled by the user such that they can pan, tilt, rotate, and zoom in and out of the scene. Review is meant to improve the ability of researchers to share their evolved results with one another. BibTeX @inproceedings{Clark.2018.GECCO.ReviewWebbasedSimulation, abstract = "Evolutionary robotics researchers often need to share results that may be too difficult to describe in text and too complex to show using images. Many researchers include links to videos as supplementary materials, but videos have a predefined view of the scene and do not allow watchers to adjust the viewing angle to their preference. In this paper we present a web-based application (based on three.js) for sharing interactive animations. Specifically, our tool (called Review) enables researchers to generate simple animation log data that can be loaded in any modern web browser on a computer or mobile device. The camera in these animations is controlled by the user such that they can pan, tilt, rotate, and zoom in and out of the scene. Review is meant to improve the ability of researchers to share their evolved results with one another.", author = "Clark, Anthony J. and Moore, Jared M.", location = "Kyoto, Japan", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2018-07-15", doi = "10.1145/3205651.3208292", eventtitle = "{GECCO} 2018", isbn = "978-1-4503-5764-7", pages = "1357--1362", shorttitle = "Review", title = "Review: A Web-Based Simulation Viewer for Sharing Evolutionary Robotics Results", }
Jul 2018

Bend and Flex: Passive Flexibility or Active Control in a Quadruped Animat

Jared M. Moore and Anthony J. Clark

Genetic and Evolutionary Computation Conference (GECCO 2018), Kyoto, Japan. DOI: 10.1145/3205651.3205703

PDF DOI Abstract Abstract: Muscle and tendon elasticity enables animals to interact with their environment softly, reducing ground impact force and increasing efficiency of locomotion. Traditional rigid body robots remain the commercially viable option, but incorporating flexibility can harness the benefits exhibited by natural organisms. In this paper, we examine how the addition of passive flexibility impacts performance and locomotive efficiency in a quadruped animat. Results show that the addition of flexibility in the spine and lower limbs of a quadruped animat significantly increases the distance traveled compared to a fully rigid-body animat. However, replacing these passively flexibile joints with actively controlled joints results in the farthest traveling individuals while maintaining similar efficiency. It appears that increases in DOF and joint configuration are the drivers of performance increases rather than passive flexibility. BibTeX @inproceedings{Moore.2018.GECCO.BendFlexPassive, abstract = "Muscle and tendon elasticity enables animals to interact with their environment softly, reducing ground impact force and increasing efficiency of locomotion. Traditional rigid body robots remain the commercially viable option, but incorporating flexibility can harness the benefits exhibited by natural organisms. In this paper, we examine how the addition of passive flexibility impacts performance and locomotive efficiency in a quadruped animat. Results show that the addition of flexibility in the spine and lower limbs of a quadruped animat significantly increases the distance traveled compared to a fully rigid-body animat. However, replacing these passively flexibile joints with actively controlled joints results in the farthest traveling individuals while maintaining similar efficiency. It appears that increases in DOF and joint configuration are the drivers of performance increases rather than passive flexibility.", author = "Moore, Jared M. and Clark, Anthony J.", location = "Kyoto, Japan", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2018-07-15", doi = "10.1145/3205651.3205703", eventtitle = "{GECCO} 2018", isbn = "978-1-4503-5764-7", shorttitle = "Bend and Flex", title = "Bend and Flex: Passive Flexibility or Active Control in a Quadruped Animat", }
Jul 2018

Evo-ROS: Integrating Evolution and the Robot Operating System

Glen A. Simon, Jared M. Moore, Anthony J. Clark, and Philip K. McKinley

Genetic and Evolutionary Computation Conference (GECCO 2018), Kyoto, Japan. DOI: 10.1145/3205651.3208269

PDF DOI Abstract Abstract: In this paper, we describe the Evo-ROS framework, which is intended to help bridge the gap between the evolutionary and traditional robotics communities. Evo-ROS combines an evolutionary algorithm with individual physics-based evaluations conducted using the Robot Operating System (ROS) and the Gazebo simulation environment. Our goals in developing Evo-ROS are to (1) provide researchers in evolutionary robotics with access to the extensive support for real-world components and capabilities developed by the ROS community and (2) enable ROS developers, and more broadly robotics researchers, to take advantage of evolutionary search during design and testing. We describe the details of the Evo-ROS structure and operation, followed by presentation of a case study using Evo-ROS to optimize placement of sonar sensors on unmanned ground vehicles that can experience reduced sensing capability due to component failures and physical damage. The case study provides insights into the current capabilities and identifies areas for future enhancements. BibTeX @inproceedings{Simon.2018.GECCO.EvoROSIntegratingEvolution, abstract = "In this paper, we describe the Evo-ROS framework, which is intended to help bridge the gap between the evolutionary and traditional robotics communities. Evo-ROS combines an evolutionary algorithm with individual physics-based evaluations conducted using the Robot Operating System (ROS) and the Gazebo simulation environment. Our goals in developing Evo-ROS are to (1) provide researchers in evolutionary robotics with access to the extensive support for real-world components and capabilities developed by the ROS community and (2) enable ROS developers, and more broadly robotics researchers, to take advantage of evolutionary search during design and testing. We describe the details of the Evo-ROS structure and operation, followed by presentation of a case study using Evo-ROS to optimize placement of sonar sensors on unmanned ground vehicles that can experience reduced sensing capability due to component failures and physical damage. The case study provides insights into the current capabilities and identifies areas for future enhancements.", author = "Simon, Glen A. and Moore, Jared M. and Clark, Anthony J. and McKinley, Philip K.", location = "Kyoto, Japan", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2018-07-15", doi = "10.1145/3205651.3208269", eventtitle = "{GECCO} 2018", isbn = "978-1-4503-5764-7", pages = "1386--1393", shorttitle = "Evo-{ROS}", title = "Evo-{ROS}: Integrating Evolution and the Robot Operating System", }
Dec 2017

Evolving Adabot: A Mobile Robot With Adjustable Wheel Extensions

Anthony J. Clark

IEEE Symposium Series on Computational Intelligence (RiiSS 2017), Honolulu, Hawaii, USA. DOI: 10.1109/SSCI.2017.8280979

PDF DOI Slides Abstract Abstract: Robots are increasingly being utilized in unstructured environments. Autonomous mobile robots are being assigned with tasks that are either too difficult or too dangerous for people. For instance, search and rescue robots can be deployed in unstable environments to aid in the search for disaster victims. In this paper, we propose a novel design for an autonomous mobile robot that can dynamically adjust traction during runtime. Our device, called Adabot meaning adaptive robot, is small, has a simple design, and can extend wegs from its wheels by adjustable amounts. We optimize both the morphology and the control parameters of Adabot using differential evolution, and our simulation results show that Adabot is effectively able to take advantage of both purely wheeled locomotion and legged-wheel locomotion by transitioning automatically between these two modes. BibTeX @inproceedings{Clark.2017.SSCI.EvolvingAdabotMobile, abstract = "Robots are increasingly being utilized in unstructured environments. Autonomous mobile robots are being assigned with tasks that are either too difficult or too dangerous for people. For instance, search and rescue robots can be deployed in unstable environments to aid in the search for disaster victims. In this paper, we propose a novel design for an autonomous mobile robot that can dynamically adjust traction during runtime. Our device, called Adabot meaning adaptive robot, is small, has a simple design, and can extend wegs from its wheels by adjustable amounts. We optimize both the morphology and the control parameters of Adabot using differential evolution, and our simulation results show that Adabot is effectively able to take advantage of both purely wheeled locomotion and legged-wheel locomotion by transitioning automatically between these two modes.", author = "Clark, Anthony J.", location = "Honolulu, Hawaii, USA", publisher = "{IEEE}", booktitle = "{IEEE} Symposium Series on Computational Intelligence", date = "2017-12-01", doi = "10.1109/SSCI.2017.8280979", eventtitle = "{RiiSS} 2017", isbn = "978-1-5386-2726-6", shorttitle = "Evolving Adabot", title = "Evolving Adabot: A Mobile Robot with Adjustable Wheel Extensions", }
Jul 2017

Effect of Animat Complexity on the Evolution of Hierarchical Control

Jared M. Moore, Anthony J. Clark, and Philip K. McKinley

Genetic and Evolutionary Computation Conference (GECCO 2017), Berlin, Germany. DOI: 10.1145/3071178.3071246

PDF DOI Abstract Abstract: Animal movements are realized by a combination of high-level control from the nervous system and joint-level movement provided by the musculoskeletal system. The digital muscle model (DMM) emulates the low-level musculoskeletal system and can be combined with a high-level artificial neural network (ANN) controller forming a hybrid control strategy. Previous work has shown that, compared to ANN-only controllers, hybrid ANN/DMM controllers exhibit similar performance with fewer synapses, suggesting that some computation is offloaded to the low-level DMM. An open question is how the complexity of the robot, in terms of the number of joints, affects the evolution of the ANN control structure. We explore this question by evolving both hybrid controllers and ANN-only controllers for worm-like animats of varying complexity. Specifically, the number of joints in the worms ranges from 1 to 12. Consistent with an earlier study, the results demonstrate that, in most cases, hybrid ANN/DMM controllers exhibit equal or better performance than ANN-only controllers. In addition, above a threshold for animat complexity (number of joints), the ANNs for one variant of the hybrid controllers have significantly fewer connections than the ANN-only controllers. BibTeX @inproceedings{Moore.2017.GECCO.EffectAnimatComplexity, abstract = "Animal movements are realized by a combination of high-level control from the nervous system and joint-level movement provided by the musculoskeletal system. The digital muscle model (DMM) emulates the low-level musculoskeletal system and can be combined with a high-level artificial neural network (ANN) controller forming a hybrid control strategy. Previous work has shown that, compared to ANN-only controllers, hybrid ANN/DMM controllers exhibit similar performance with fewer synapses, suggesting that some computation is offloaded to the low-level DMM. An open question is how the complexity of the robot, in terms of the number of joints, affects the evolution of the ANN control structure. We explore this question by evolving both hybrid controllers and ANN-only controllers for worm-like animats of varying complexity. Specifically, the number of joints in the worms ranges from 1 to 12. Consistent with an earlier study, the results demonstrate that, in most cases, hybrid ANN/DMM controllers exhibit equal or better performance than ANN-only controllers. In addition, above a threshold for animat complexity (number of joints), the ANNs for one variant of the hybrid controllers have significantly fewer connections than the ANN-only controllers.", author = "Moore, Jared M. and Clark, Anthony J. and McKinley, Philip K.", location = "Berlin, Germany", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2017-07-15", doi = "10.1145/3071178.3071246", eventtitle = "{GECCO} 2017", isbn = "978-1-4503-4920-8", pages = "147--154", title = "Effect of Animat Complexity on the Evolution of Hierarchical Control", }
Dec 2016

An Evolutionary Approach to Discovering Execution Mode Boundaries for Adaptive Controllers

Anthony J. Clark, Byron DeVries, Jared M. Moore, Betty H. C. Cheng, and Philip K. McKinley

IEEE Symposium Series on Computational Intelligence (SSCI 2016), Athens, Greece. DOI: 10.1109/SSCI.2016.7850178

PDF DOI Slides Abstract Abstract: Adaptive controllers enable cyberphysical systems, such as autonomous robots, to manage uncertain conditions during execution. However, there is a limit to the range of conditions that can be handled by a given controller. When this limit is exceeded, a controller might fail to respond as expected, not only rendering it ineffective but possibly putting the entire system at risk. In this paper, we describe a method based on evolutionary search for automatically enhancing, and discovering the boundaries of, a given adaptive controller. Collectively, these boundaries define an execution mode for that controller. Explicit specification of mode boundaries facilitates the development of decision logic that determines, based on system state and sensed conditions, when to switch to a different execution mode and typically a different controller, such as one for providing fail-safe operation. To evaluate the proposed approach, we apply it to a robotic fish propelled by a flexible caudal fin that is governed by a model-free adaptive controller. Experimental results demonstrate that this approach is effective in characterizing a controller’s ability to adapt to environmental dynamics, including physical damage to the robot itself. BibTeX @inproceedings{Clark.2016.ICES.EvolutionaryApproachDiscovering, abstract = "Adaptive controllers enable cyberphysical systems, such as autonomous robots, to manage uncertain conditions during execution. However, there is a limit to the range of conditions that can be handled by a given controller. When this limit is exceeded, a controller might fail to respond as expected, not only rendering it ineffective but possibly putting the entire system at risk. In this paper, we describe a method based on evolutionary search for automatically enhancing, and discovering the boundaries of, a given adaptive controller. Collectively, these boundaries define an execution mode for that controller. Explicit specification of mode boundaries facilitates the development of decision logic that determines, based on system state and sensed conditions, when to switch to a different execution mode and typically a different controller, such as one for providing fail-safe operation. To evaluate the proposed approach, we apply it to a robotic fish propelled by a flexible caudal fin that is governed by a model-free adaptive controller. Experimental results demonstrate that this approach is effective in characterizing a controller’s ability to adapt to environmental dynamics, including physical damage to the robot itself.", author = "Clark, Anthony J. and DeVries, Byron and Moore, Jared M. and Cheng, Betty H. C. and McKinley, Philip K.", location = "Athens, Greece", publisher = "{IEEE}", booktitle = "{IEEE} Symposium Series on Computational Intelligence", date = "2016-12-15", doi = "10.1109/SSCI.2016.7850178", eventtitle = "{SSCI} 2016", isbn = "978-1-5090-4240-1", title = "An Evolutionary Approach to Discovering Execution Mode Boundaries for Adaptive Controllers", }
Nov 2015

Evolutionary Multiobjective Design of a Flexible Caudal Fin for Robotic Fish

Anthony J. Clark, Xiaobo Tan, and Philip K. McKinley

Bioinspiration & Biomimetics. DOI: 10.1088/1748-3190/10/6/065006

PDF DOI Abstract Abstract: Robotic fish accomplish swimming by deforming their bodies or other fin-like appendages. As an emerging class of embedded computing system, robotic fish are anticipated to play an important role in environmental monitoring, inspection of underwater structures, tracking of hazardous wastes and oil spills, and the study of live fish behaviors. While integration of flexible materials (into the fins and/or body) holds the promise of improved swimming performance (in terms of both speed and maneuverability) for these robots, such components also introduce significant design challenges due to the complex material mechanics and hydrodynamic interactions. The problem is further exacerbated by the need for the robots to meet multiple objectives (e.g., both speed and energy efficiency). In this paper, we propose an evolutionary multiobjective optimization approach to the design and control of a robotic fish with a flexible caudal fin. Specifically, we use the NSGA-II algorithm to investigate morphological and control parameter values that optimize swimming speed and power usage. Several evolved fin designs are validated experimentally with a small robotic fish, where fins of different stiffness values and sizes are printed with a multi-material 3D printer. Experimental results confirm the effectiveness of the proposed design approach in balancing the two competing objectives. BibTeX @article{Clark.2015.BB.EvolutionaryMultiobjectiveDesign, abstract = "Robotic fish accomplish swimming by deforming their bodies or other fin-like appendages. As an emerging class of embedded computing system, robotic fish are anticipated to play an important role in environmental monitoring, inspection of underwater structures, tracking of hazardous wastes and oil spills, and the study of live fish behaviors. While integration of flexible materials (into the fins and/or body) holds the promise of improved swimming performance (in terms of both speed and maneuverability) for these robots, such components also introduce significant design challenges due to the complex material mechanics and hydrodynamic interactions. The problem is further exacerbated by the need for the robots to meet multiple objectives (e.g., both speed and energy efficiency). In this paper, we propose an evolutionary multiobjective optimization approach to the design and control of a robotic fish with a flexible caudal fin. Specifically, we use the NSGA-II algorithm to investigate morphological and control parameter values that optimize swimming speed and power usage. Several evolved fin designs are validated experimentally with a small robotic fish, where fins of different stiffness values and sizes are printed with a multi-material 3D printer. Experimental results confirm the effectiveness of the proposed design approach in balancing the two competing objectives.", author = "Clark, Anthony J. and Tan, Xiaobo and McKinley, Philip K.", date = "2015-11-25", doi = "10.1088/1748-3190/10/6/065006", issn = "1748-3190", journaltitle = "Bioinspiration \& Biomimetics", number = "6", title = "Evolutionary Multiobjective Design of a Flexible Caudal Fin for Robotic Fish", volume = "10", }
Jul 2015

Enhancing a Model-Free Adaptive Controller Through Evolutionary Computation

Anthony J. Clark, Philip K. McKinley, and Xiaobo Tan

Genetic and Evolutionary Computation Conference (GECCO 2015), Madrid, Spain. DOI: 10.1145/2739480.2754762

PDF DOI Slides Abstract Abstract: Many robotic systems experience fluctuating dynamics during their lifetime. Variations can be attributed in part to material degradation and decay of mechanical hardware. One approach to mitigating these problems is to utilize an adaptive controller. For example, in model-free adaptive control (MFAC) a controller learns how to drive a system by continually updating link weights of an artificial neural network (ANN). However, determining the optimal control parameters for MFAC, including the structure of the underlying ANN, is a challenging process. In this paper we investigate how to enhance the online adaptability of MFAC-based systems through computational evolution. We apply the proposed methods to a simulated robotic fish propelled by a flexible caudal fin. Results demonstrate that the robot is able to effectively respond to changing fin characteristics and varying control signals when using an evolved MFAC controller. Notably, the system is able to adapt to characteristics not encountered during evolution. The proposed technique is general and can be applied to improve the adaptability of other cyber-physical systems. BibTeX @inproceedings{Clark.2015.GECCO.EnhancingModelFreeAdaptive, abstract = "Many robotic systems experience fluctuating dynamics during their lifetime. Variations can be attributed in part to material degradation and decay of mechanical hardware. One approach to mitigating these problems is to utilize an adaptive controller. For example, in model-free adaptive control (MFAC) a controller learns how to drive a system by continually updating link weights of an artificial neural network (ANN). However, determining the optimal control parameters for MFAC, including the structure of the underlying ANN, is a challenging process. In this paper we investigate how to enhance the online adaptability of MFAC-based systems through computational evolution. We apply the proposed methods to a simulated robotic fish propelled by a flexible caudal fin. Results demonstrate that the robot is able to effectively respond to changing fin characteristics and varying control signals when using an evolved MFAC controller. Notably, the system is able to adapt to characteristics not encountered during evolution. The proposed technique is general and can be applied to improve the adaptability of other cyber-physical systems.", author = "Clark, Anthony J. and McKinley, Philip K. and Tan, Xiaobo", location = "Madrid, Spain", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2015-07-12", doi = "10.1145/2739480.2754762", eventtitle = "{GECCO} 2015", isbn = "978-1-4503-3472-3", langid = "english", pages = "137--144", title = "Enhancing a Model-Free Adaptive Controller through Evolutionary Computation", }
Dec 2014

Balancing Performance and Efficiency in a Robotic Fish With Evolutionary Multiobjective Optimization

Anthony J. Clark, Jianxun Wang, Xiaobo Tan, and Philip K. McKinley

IEEE International Conference on Evolvable Systems (ICES 2014), Orlando, Florida, USA. DOI: 10.1109/ICES.2014.7008744

PDF DOI Slides Abstract Abstract: In this paper, we apply evolutionary multiobjective optimization to the design of a robotic fish with a flexible caudal fin. Specifically, we use the NSGA-II algorithm to discover solutions (physical dimensions, flexibility, and control parameters) that optimize both swimming performance and power efficiency. The optimization is conducted in a custom simulation environment based on an accurate yet computationally-efficient model of hydrodynamics. The results of these simulations reveal general principles that can be applied in the design of robotic fish morphology and control. To verify that the simulation results are physically relevant, we selected several of the evolved solutions, fabricated flexible caudal fins using a multi-material 3D printer, and attached them to a robotic fish prototype. Experimental results, conducted in a large water tank, correspond reasonably well to simulation results in both swimming performance and power efficiency, demonstrating the usefulness of evolutionary computation methods to this application domain. BibTeX @inproceedings{Clark.2014.ICES.BalancingPerformanceEfficiency, abstract = "In this paper, we apply evolutionary multiobjective optimization to the design of a robotic fish with a flexible caudal fin. Specifically, we use the NSGA-II algorithm to discover solutions (physical dimensions, flexibility, and control parameters) that optimize both swimming performance and power efficiency. The optimization is conducted in a custom simulation environment based on an accurate yet computationally-efficient model of hydrodynamics. The results of these simulations reveal general principles that can be applied in the design of robotic fish morphology and control. To verify that the simulation results are physically relevant, we selected several of the evolved solutions, fabricated flexible caudal fins using a multi-material 3D printer, and attached them to a robotic fish prototype. Experimental results, conducted in a large water tank, correspond reasonably well to simulation results in both swimming performance and power efficiency, demonstrating the usefulness of evolutionary computation methods to this application domain.", author = "Clark, Anthony J. and Wang, Jianxun and Tan, Xiaobo and McKinley, Philip K.", location = "Orlando, Florida, USA", publisher = "{IEEE}", booktitle = "{IEEE} International Conference on Evolvable Systems", date = "2014-12-15", doi = "10.1109/ICES.2014.7008744", eventtitle = "{ICES} 2014", isbn = "978-1-4799-4479-8", pages = "227--234", title = "Balancing Performance and Efficiency in a Robotic Fish with Evolutionary Multiobjective Optimization", }
Jul 2014

On-Board Evolution of a Model-Free Adaptive Controller for a Robotic Fish

Anthony J. Clark, Philip K. McKinley, and Xiaobo Tan

Evolution of Physical Systems Workshop, Held in Conjunction With the International Conference on the Synthesis and Simulation of Living Systems (ALIFE 2014), New York City, New York, USA.

PDF Slides Abstract Abstract: Many physical systems experience fluctuating dynamics throughout their lifetime. Variations can be attributed in part to material degradation and decay of mechanical hardware. Designing control strategies that mitigate the negative effects of such variations can be difficult. One approach is to utilize model-free adaptive control (MFAC), which learns how to control a system by continually updating link weights of an artificial neural network (ANN) (Cheng, 2004). However, determining the optimal values of various control parameters, as well as the structure of the ANN, is challenging. In this study, we investigate how to enhance the on-board adaptability of MFAC-based systems through computational evolution. BibTeX @inproceedings{Clark.2014.ALIFE.OnBoardEvolutionModelFree, abstract = "Many physical systems experience fluctuating dynamics throughout their lifetime. Variations can be attributed in part to material degradation and decay of mechanical hardware. Designing control strategies that mitigate the negative effects of such variations can be difficult. One approach is to utilize model-free adaptive control (MFAC), which learns how to control a system by continually updating link weights of an artificial neural network (ANN) (Cheng, 2004). However, determining the optimal values of various control parameters, as well as the structure of the ANN, is challenging. In this study, we investigate how to enhance the on-board adaptability of MFAC-based systems through computational evolution.", author = "Clark, Anthony J. and McKinley, Philip K. and Tan, Xiaobo", location = "New York City, New York, USA", booktitle = "Evolution of Physical Systems Workshop, Held in Conjunction with the International Conference on the Synthesis and Simulation of Living Systems", date = "2014-07-30", eventtitle = "{ALIFE} 2014", title = "On-Board Evolution of a Model-Free Adaptive Controller for a Robotic Fish", }
Jul 2014

Evolutionary Robotics on the Web With WebGL and JavaScript

Jared M. Moore, Anthony J. Clark, and Philip K. McKinley

International Conference on the Synthesis and Simulation of Living Systems (ALIFE 2014), New York City, New York, USA.

PDF Abstract Abstract: Web-based applications are highly accessible to users, providing rich, interactive content while eliminating the need to install software locally. However, evolutionary robotics (ER) has faced challenges in this domain as web-based technologies have not been amenable to 3D physics simulations. Traditionally, physics-based simulations require a local installation and a high degree of user knowledge to configure an environment, but the emergence of Javascript-based physics engines enables complex simulations to be executed in web browsers. These developments create opportunities for ER research to reach new audiences by increasing accessibility. In this work, we introduce two web-based tools we have built to facilitate the exchange of ideas with other researchers as well as outreach to K-12 students and the general public. The first tool is intended to distribute and exchange ER research results, while the second is a completely browser-based implementation of an ER environment. BibTeX @inproceedings{Moore.2014.ALIFE.EvolutionaryRoboticsWeb, abstract = "Web-based applications are highly accessible to users, providing rich, interactive content while eliminating the need to install software locally. However, evolutionary robotics (ER) has faced challenges in this domain as web-based technologies have not been amenable to 3D physics simulations. Traditionally, physics-based simulations require a local installation and a high degree of user knowledge to configure an environment, but the emergence of Javascript-based physics engines enables complex simulations to be executed in web browsers. These developments create opportunities for ER research to reach new audiences by increasing accessibility. In this work, we introduce two web-based tools we have built to facilitate the exchange of ideas with other researchers as well as outreach to K-12 students and the general public. The first tool is intended to distribute and exchange ER research results, while the second is a completely browser-based implementation of an ER environment.", author = "Moore, Jared M. and Clark, Anthony J. and McKinley, Philip K.", location = "New York City, New York, USA", url = "http://arxiv.org/abs/1406.3337", booktitle = "International Conference on the Synthesis and Simulation of Living Systems", date = "2014-07-30", eprint = "1406.3337", eprinttype = "arxiv", eventtitle = "{ALIFE} 2014", title = "Evolutionary Robotics on the Web with {WebGL} and {JavaScript}", }
Jul 2014

Hold the Spot: Evolution of Generalized Station Keeping for an Aquatic Robot

Jared M. Moore and Anthony J. Clark

International Conference on the Synthesis and Simulation of Living Systems (ALIFE 2014), New York City, New York, USA. DOI: 10.7551/978-0-262-32621-6-ch033

PDF DOI Abstract Abstract: In this paper, we present a strategy to evolve neurocontrollers in aquatic robots capable of generalized station keeping, that is, maintaining a position in the presence of various water flows. Evolved behaviors exhibit a variety of complex fin/flipper movements that enable the robot to react and move against changing flows. Moreover, results indicate that some sensor modalities are beneficial when the robot is placed in novel environments, though little used during the evolutionary process. BibTeX @inproceedings{Moore.2014.ALIFE.HoldSpotEvolution, abstract = "In this paper, we present a strategy to evolve neurocontrollers in aquatic robots capable of generalized station keeping, that is, maintaining a position in the presence of various water flows. Evolved behaviors exhibit a variety of complex fin/flipper movements that enable the robot to react and move against changing flows. Moreover, results indicate that some sensor modalities are beneficial when the robot is placed in novel environments, though little used during the evolutionary process.", author = "Moore, Jared M. and Clark, Anthony J.", location = "New York City, New York, USA", publisher = "The MIT Press", booktitle = "International Conference on the Synthesis and Simulation of Living Systems", date = "2014-07-30", doi = "10.7551/978-0-262-32621-6-ch033", eventtitle = "{ALIFE} 2014", isbn = "978-0-262-32621-6", shorttitle = "Hold the Spot", title = "Hold the Spot: Evolution of Generalized Station Keeping for an Aquatic Robot", }
Sep 2013

Just Keep Swimming: Accounting for Uncertainty in Self-Modeling Aquatic Robots

Matthew J. Rose, Anthony J. Clark, Jared M. Moore, and Philip K. McKinley

International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems (ERLARS 2013), Taormina, Italy. Best Paper Award

PDF Abstract Abstract: A robust robotic system should be able to overcome unforeseen conditions, including physical damage and component failure occurring after deployment. A self-modeling system maintains an internal image of itself, which can be updated to reflect incurred damage. The robot can use this model to derive (or evolve) new behaviors such as gaits that account for the damage. In this paper we describe an approach to self-modeling for aquatic robots. The aquatic environment presents unique challenges to the self-modeling process, including the inherent uncertainty in the robot’s orientation and configuration. We propose and evaluate two approaches to automatically infer missing contextual information, which otherwise complicates the task of developing an accurate model. We demonstrate the effectiveness of these methods on a particular aquatic robot intended for remote sensing. BibTeX @inproceedings{Rose.2013.ERLARS.JustKeepSwimming, abstract = "A robust robotic system should be able to overcome unforeseen conditions, including physical damage and component failure occurring after deployment. A self-modeling system maintains an internal image of itself, which can be updated to reflect incurred damage. The robot can use this model to derive (or evolve) new behaviors such as gaits that account for the damage. In this paper we describe an approach to self-modeling for aquatic robots. The aquatic environment presents unique challenges to the self-modeling process, including the inherent uncertainty in the robot’s orientation and configuration. We propose and evaluate two approaches to automatically infer missing contextual information, which otherwise complicates the task of developing an accurate model. We demonstrate the effectiveness of these methods on a particular aquatic robot intended for remote sensing.", author = "Rose, Matthew J. and Clark, Anthony J. and Moore, Jared M. and McKinley, Philip K.", location = "Taormina, Italy", booktitle = "International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems", date = "2013-09-15", eventtitle = "{ERLARS} 2013", shorttitle = "Just Keep Swimming", title = "Just Keep Swimming: Accounting for Uncertainty in Self-Modeling Aquatic Robots", note = "Best Paper Award", }
Jul 2013

Evolutionary Optimization of Robotic Fish Control and Morphology

Anthony J. Clark and Philip K. McKinley

Genetic and Evolutionary Computation Conference (GECCO 2013), Amsterdam, The Netherlands. DOI: 10.1145/2464576.2464593

PDF DOI Slides Abstract Abstract: The nonlinear dynamics of an aquatic environment make robotic fish behavior difficult to predict and subsequently difficult to optimize. In this paper, we present a method for optimizing robotic fish propulsion through the evolution of control patterns and caudal fin flexibility. Evolved solutions are evaluated in a physics-based simulation environment. Control signals are generated with both simple sinusoids and neural oscillators. This study demonstrates how evolutionary algorithms can be utilized to handle the complex interactions among material properties, physical form, and control patterns in an aquatic environment. BibTeX @inproceedings{Clark.2013.GECCO.EvolutionaryOptimizationRobotic, abstract = "The nonlinear dynamics of an aquatic environment make robotic fish behavior difficult to predict and subsequently difficult to optimize. In this paper, we present a method for optimizing robotic fish propulsion through the evolution of control patterns and caudal fin flexibility. Evolved solutions are evaluated in a physics-based simulation environment. Control signals are generated with both simple sinusoids and neural oscillators. This study demonstrates how evolutionary algorithms can be utilized to handle the complex interactions among material properties, physical form, and control patterns in an aquatic environment.", author = "Clark, Anthony J. and McKinley, Philip K.", location = "Amsterdam, The Netherlands", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2013-07-15", doi = "10.1145/2464576.2464593", eventtitle = "{GECCO} 2013", isbn = "978-1-4503-1964-5", langid = "english", title = "Evolutionary Optimization of Robotic Fish Control and Morphology", }
Jul 2013

Evolution of Station Keeping as a Response to Flows in an Aquatic Robot

Jared M. Moore, Anthony J. Clark, and Philip K. McKinley

Genetic and Evolutionary Computation Conference (GECCO 2013), Amsterdam, The Netherlands. DOI: 10.1145/2463372.2463402

PDF DOI Abstract Abstract: Developing complex behaviors for aquatic robots is a difficult engineering challenge due to the uncertainty of an underwater environment. Neuroevolution provides one method of dealing with this type of problem. Artificial neural networks discern different conditions by mapping sensory input to responses, and evolutionary computation provides a training algorithm suitable to the high dimensionality of the problem. In this paper, we present results of applying neuroevolution to an aquatic robot tasked with station keeping, that is, maintaining a given position despite surrounding water flow. The virtual device exposed to evolution is modeled after a physical counterpart that has been fabricated with a 3D printer and tested in physical environments. Evolved behaviors exhibit a variety of unexpected, complex fin/flipper movements that enable the robot to achieve and maintain station, despite water flow from different directions. Moreover, the results show that evolved controllers are able to effectively carry out this task using only information from a simulated accelerometer and gyroscope, matching the inertial measurement unit (IMU) on the actual robot. BibTeX @inproceedings{Moore.2013.GECCO.EvolutionStationKeeping, abstract = "Developing complex behaviors for aquatic robots is a difficult engineering challenge due to the uncertainty of an underwater environment. Neuroevolution provides one method of dealing with this type of problem. Artificial neural networks discern different conditions by mapping sensory input to responses, and evolutionary computation provides a training algorithm suitable to the high dimensionality of the problem. In this paper, we present results of applying neuroevolution to an aquatic robot tasked with station keeping, that is, maintaining a given position despite surrounding water flow. The virtual device exposed to evolution is modeled after a physical counterpart that has been fabricated with a 3D printer and tested in physical environments. Evolved behaviors exhibit a variety of unexpected, complex fin/flipper movements that enable the robot to achieve and maintain station, despite water flow from different directions. Moreover, the results show that evolved controllers are able to effectively carry out this task using only information from a simulated accelerometer and gyroscope, matching the inertial measurement unit (IMU) on the actual robot.", author = "Moore, Jared M. and Clark, Anthony J. and McKinley, Philip K.", location = "Amsterdam, The Netherlands", publisher = "ACM Press", booktitle = "Genetic and Evolutionary Computation Conference", date = "2013-07-15", doi = "10.1145/2463372.2463402", eventtitle = "{GECCO} 2013", isbn = "978-1-4503-1963-8", langid = "english", title = "Evolution of Station Keeping as a Response to Flows in an Aquatic Robot", }
Jul 2012

Evolutionary Design and Experimental Validation of a Flexible Caudal Fin for Robotic Fish

Anthony J. Clark, Jared M. Moore, Jianxun Wang, Xiaobo Tan, and Philip K. McKinley

International Conference on the Synthesis and Simulation of Living Systems (ALIFE 2013), East Lansing, Michigan, USA. Best Paper Award DOI: 10.7551/978-0-262-31050-5-ch043

PDF DOI Slides Abstract Abstract: Designing a robotic fish is a challenging endeavor due to the non-linear dynamics of underwater environments. In this paper, we present an evolutionary computation approach for designing the caudal fin of a carangiform robotic fish. Evolutionary experiments are performed in a simulated environment utilizing a mathematical model to approximate the hydrodynamic motion of a flexible caudal fin. With this model, time-consuming computational fluid dynamic simulations can be avoided while maintaining a physically realistic simulation. Two approaches are employed to maximize a robotic fish’s average velocity. First, a hill-climbing algorithm is applied to find the optimal stiffness for a fixed shape caudal fin. Next, both fin stiffness and shape are simultaneously optimized with a genetic algorithm. Additionally, simulated caudal fins are compared to physically validated fins, which were fabricated with the aid of a 3D printer and tested on a robotic fish prototype. Results show a correlation between evolved results, model predicted behavior, and physical robot performance with some disparity due to the difficulty in accurately approximating real world performance in a simulation environment. Despite the disparity, evolutionary design is shown to be a viable process. BibTeX @inproceedings{Clark.2012.ALIFE.EvolutionaryDesignExperimental, abstract = "Designing a robotic fish is a challenging endeavor due to the non-linear dynamics of underwater environments. In this paper, we present an evolutionary computation approach for designing the caudal fin of a carangiform robotic fish. Evolutionary experiments are performed in a simulated environment utilizing a mathematical model to approximate the hydrodynamic motion of a flexible caudal fin. With this model, time-consuming computational fluid dynamic simulations can be avoided while maintaining a physically realistic simulation. Two approaches are employed to maximize a robotic fish’s average velocity. First, a hill-climbing algorithm is applied to find the optimal stiffness for a fixed shape caudal fin. Next, both fin stiffness and shape are simultaneously optimized with a genetic algorithm. Additionally, simulated caudal fins are compared to physically validated fins, which were fabricated with the aid of a 3D printer and tested on a robotic fish prototype. Results show a correlation between evolved results, model predicted behavior, and physical robot performance with some disparity due to the difficulty in accurately approximating real world performance in a simulation environment. Despite the disparity, evolutionary design is shown to be a viable process.", author = "Clark, Anthony J. and Moore, Jared M. and Wang, Jianxun and Tan, Xiaobo and McKinley, Philip K.", location = "East Lansing, Michigan, USA", publisher = "MIT Press", booktitle = "International Conference on the Synthesis and Simulation of Living Systems", date = "2012-07-02", doi = "10.7551/978-0-262-31050-5-ch043", eventtitle = "{ALIFE} 2013", isbn = "978-0-262-31050-5", pages = "325--332", title = "Evolutionary Design and Experimental Validation of a Flexible Caudal Fin for Robotic Fish", note = "Best Paper Award", }

University Service
Aug 2024 to present
Information Technology Committee

Review and recommend policies and procedures for the use of information technology resources.

Aug 2022 to May 2023
Teaching and Learning Committee

Promote student learning and achievement by sustaining faculty in their development as teachers.

Jul 2021 to present
Pomona Scholars of Math (PSM) Advisor

Mentor and advise first year students.

Jul 2021 to May 2022
Fulbright Advisor

Advise students applying for Fulbright Awards.

Nov 2017 to May 2020
Robotics Club Advisor

Given strong student demand, I (and one of our EE faculty) initiated Missouri State University’s first robotics club.

Aug 2017 to May 2020
CSC Representative, CNAS College Council (elected)

Act upon curricular matters that are referred to it by departments within the college. The College Council approves departmental proposals, rejects and returns proposals to the originating department, or amends and approves proposals.

Aug 2016 to May 2020
CSC Representative, CNAS Student Recruitment Committee

Attend recruitment events on the behalf of the college, and make recommendations to the dean regarding recruitment procedures.

Apr 2018 to May 2019
STEMentors Program Advisor

Advise the new outreach program, which is directed at providing mentoring for local lower-income schools.

Aug 2016 to Aug 2019
CSC Representative, CNAS Diversity Committee

Represent my department at the college level diversity committee. A primary goal for the members of this committee is to improve the retention of students that are considered at risk for either dropping out or transferring. We improve retention through a variety of activities: poster sessions, scholarships, and picnics.

Aug 2017 to May 2019
ACM Chapter Advisor

Coordinate ACM study chapter activities, include: scheduling speakers, organizing off-campus activities (e.g., competitions), and recruit volunteers to help at departmental events.

Sep 2018
Proactive Advisor

Attend special training sessions on proactive advising techniques so that I can better advise first generation computer science undergraduates. I currently advise ~75 CS students.

Aug 2014 to May 2016
Coordinator, Computer Science and Engineering Graduate Association (elected)

Coordinated monthly meetings for graduate students in the Department of Computer Science and Engineering, facilitated communication of Department news and policies, and organized graduate student service opportunities.

Aug 2014 to May 2015
Graduate Representative, Computer Science and Engineering Graduate Studies and Research Committee (elected)

Act as a voting member of the GSRC, which establishes academic standards, coordinates graduate course offerings, determines admission standards and policies for financial awards, and evaluates Ph.D. qualifier examinations.

Aug 2013 to May 2014
Graduate Representative, Computer Science and Engineering Departmental Meetings (elected)

Act as a voting member at CSE department meetings.

Aug 2007 to May 2009
Officer, Eta Kappa Nu, Electrical and Computer Engineering at Kansas State University (elected)

Professional Activities
Reviewer for Journals

Applied Sciences
Elsevier Robotics and Autonomous Systems
IEEE Transactions on Systems, Man and Cybernetics: Systems
IEEE Transactions on Robotics
Sage Adaptive Behavior
Sage International Journal of Advanced Robotic Systems

Professional Society Memberships

IEEE and ACM

NSF Panelist

Smart and Autonomous Systems
National Robotics Initiative
IIS Robust Intelligence

Jul 2024
Tutorial: Simulation in Evolutionary Robotics (SimER)
ALIFE 2024 Conference

Led a tutorial in the use of simulation in evolutionary robotics.

Jan 2018 to Jan 2021
Task Force Member
IEEE Task Force on Evolutionary Developmental Systems and Robotics
Apr 2022
Program Committee Member
EvoStar Conference
Brno, Czech Republic
Dec 2021
Program Committee Member
IEEE Symposium Series on Computational Intelligence
(Remote) Orlando, Florida, USA
Apr 2021
Reviewer
EvoApps, International Conference on the Applications of Evolutionary Computation
Seville, Spain
Dec 2021
Reviewer
IEEE Symposium Series on Computational Intelligence
Orlando, Florida, USA
Dec 2020
Reviewer
IEEE Symposium Series on Computational Intelligence
Canberra, Australia
Apr 2020
Program Committee Member
EvoApps, International Conference on the Applications of Evolutionary Computation
Seville, Spain
Apr 2018
Program Committee Member
EvoROBOT, European Conference on the Applications of Evolutionary Computation
Parma, Italy
Sep 2017
Abstract and Poster
IEEE/RSJ International Conference on Intelligent Robots and Systems

Evo-ROS: Integrating Evolutionary Robotics and ROS
Moore, Jared M., Clark, Anthony J., Simon, Glen, McKinley, Philip K. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems. Vancouver, BC, Canada, September, 2017

Jul 2017
Workshop Organizer, SimER: Simulation in Evolutionary Robotics Workshop
Genetic and Evolutionary Computation Conference
Berlin, Germany

Responsibilities: Co-organize a workshop that brought together experts from around the world to discuss the topic of simulation; specifically how we can improve the current state of simulation in ER.

Dec 2014
Reviewer
IEEE Symposium Series on Computational Intelligence
Orlando, Florida, USA
Sep 2014
Reviewer
IEEE International Conference on Self-Adaptive and Self-Organizing Systems
London, UK
Sep 2013
Invited Conference Talk: Evolving Aquatic Robots

The Twelfth European Conference on Artificial Life (ECAL), International Evolution of Physical Systems Workshop (EPS)
Taormina, Italy, September 2013

Sep 2013
Reviewer
IEEE International Conference on Self-Adaptive and Self-Organizing Systems
Philadelphia, Pennsylvania, USA

Advising, Mentoring, and Outreach
Nov 2017 to Jan 2020
First Lego League Team Mentor and Competition Judge
FIRST

Responsibilities: Mentor one FLL team and judge at a regional competition.

Aug 2016 to May 2020
Advising Undergraduate and Graduate Researchers
Missouri State University

Responsibilities: Coordinate and advise from three to six undergraduate and graduate students from different departments every semester.

Jun 2019
Week-long Summer Coding Camp
Discover Center and Missouri State University

Responsibilities: Organized and taught a week-long summer coding camp for middle school students.

Jul 2015
Instructor, Introduction to Evolutionary Robotics
Introduction to Robotics Engineering Program, Michigan State University

Responsibilities: Presented my research and an explanation of evolutionary robotics to 22 high school students. I introduced a web-based evolutionary robotics simulation platform (BoxCar2D) to the students and in a hands-on laboratory session helped them answer several questions regarding the evolutionary robotics process.

Mar to Jul 2015
Mentor, Visiting Scholar Program
Department of Computer Science and Engineering, Michigan State University

Responsibilities: Co-mentored Mr. René Draschwandtner, a visiting Master’s student from the University of Applied Sciences in Austria. I worked with Mr. Draschwandtner, Dr. Jared Moore, and Dr. Philip McKinley to study locomotion and grasping behaviors for a snake-like robot using methods from evolutionary robotics.

Sep 2014
Presenter, 3D Printing Showcase
Michigan State University Library

Responsibilities: Presented 3D printing technologies and my lab’s research as part of outreach directed at undergraduates and the general public.

Aug 2014
Co-Organizer of Sandbox Session, Evolution-In-Action Software and the Web
NSF BEACON Center

Responsibilities: Organized an open discussion regarding the application of state-of-the-art web technologies to evolutionary research and outreach projects.

Jul 2014
Instructor, BEACON High School Summer Residential Program
W.K. Kellogg Biological Station, Michigan State University

Responsibilities: Presented an overview of evolutionary computation to a group of four high school students interested in STEM fields, and then facilitated their work as they conducted, wrote about, and presented results from their own evolutionary study in a day-long course.

Jul 2014
Mentor, NSF Research Experience for Teachers Summer Program
College of Engineering, Michigan State University

Responsibilities: Mentored a local high school engineering instructor, Charles Payson. Over the course of his second summer in the program, Mr. Payson designed, implemented, and presented a web application used to teach evolutionary robotics concepts to K-12 students and the general public. I taught Mr. Payson web-programming skills as well as aided him in developing a curriculum for high school students.

Jul 2014
Instructor, Introduction to Robotics Engineering
College of Engineering High School Summer Program, Michigan State University

Responsibilities: Introduced evolutionary robotics to approximately 20 high school students in a tutorial style. The tutorial was based on an interactive web-based simulation environment developed by myself and Jared M. Moore. Students conducted evolutionary experiments in which they evolved robots in simulation.

Feb 2014
Graduate Student Evaluator
Undergraduate Research and Arts Forum, Michigan State University

Responsibilities: Provided feedback to undergraduates presenting their research, and scored poster presentations for a competition.

May to Aug 2013
Mentor, NSF Research Experience for Teachers Summer Program
College of Engineering, Michigan State University

Responsibilities: Mentored a local high school engineering instructor, Charles Payson. During a six-week program, I aided Mr. Payson in learning C++ programming, evolutionary algorithm development, and creating dynamic simulations. At the end of the program, I assisted Mr. Payson in translating his research into a robotics lesson plan using the VEX robotics platform.

May 2011 to Aug 2013
Mentor, NSF Research Experience for Teachers Summer Program
College of Engineering, Michigan State University

Responsibilities: Mentored a local elementary school teacher, Adam Ford, who specializes in computers and robotics. Mr. Ford developed the Biolume environment, which demonstrates evolution ‘in-action’ using using simple robots. The Biolume project is an outreach exhibit aimed at demonstrating evolutionary principles to the general public.


Funding and Grant Activity
I have contributed to writing, editing, and producing preliminary results for the following grants.
Aug 2019
Missouri Space Grant Program

Title: Lunar geologic compass for geologic mapping and surveying
Award: Student Support
Details: Awarded funds to conduct research with graduate and undergraduate students.
PIs: Matt McKay and Anthony J. Clark

May 2018
Missouri State University Summer Faculty Fellowship

Award: $6,000
Details: Funds were awarded to work on research during the summer term.
PI: Anthony J. Clark

Jul 2018
NVIDIA GPU Grant Program

Award: Quadro P6000
Details: NVIDIA awarded a powerful GPU to be used for deep learning research.
PI: Anthony J. Clark

Nov 2016
Missouri State University Major Equipment Grant

Amount: $24,000
Details: Funds were utilized to purchase a 3D printer and a CNC mill that will be used by faculty and students in the Departments of Computer Science and Engineering.
PI: Anthony J. Clark

Aug 2013 to May 2014
Distributed, Onboard Evolution in a Robotic Cloud

Amount: $168,231
Sponsor: NSF BEACON Center for the Study of Evolution in Action
PI: T. Soule (U. Idaho), Co-PIs: R. Heckendorn (U. Idaho), P. McKinley (MSU), J. Zhan (NCA&T), S. Harrison (NCA&T)

Aug 2011 to May 2014
II-EN: Evolution Park: An Evolutionary Robotics Habitat for the Study of Crawling, Swimming and Flying Creatures

Amount: $305,000
Sponsor: NSF, Division Of Computer and Network Systems
PI: P. McKinley, Co-PIs: X. Tan, J. Boughman

Aug 2012 to May 2013
Understanding and Synthesizing Collective Behavior with Mixed Robotic and Live Fish Schools

Amount: $169,923
Sponsor: NSF BEACON Center for the Study of Evolution in Action
PI: X. Tan, Co-PIs: P. McKinley, J. Boughman

Aug 2011 to May 2012
Exploiting Robot-Fish Interactions and Evolutionary Computing to Understand and Synthesize Complex Collective Behavior

Amount: $110,642
Sponsor: NSF BEACON Center for the Study of Evolution in Action
PI: X. Tan, Co-PIs: P. McKinley, J. Boughman


Software Contributions
Review, A Visualization Player

Description: Review is a web-based platform for sharing dynamic visualizations. Code, Description

Evolve-a-Robot, An Online Evolutionary Robotics Environment

Description: Evolve-a-Robot is an interactive evolutionary robotics simulation. The project has two goals. The first is to expose K-12 students to evolution and evolutionary computation using an engaging and fun platform. Evolve-a-Robot does this by visualizing the evolution of robotic cars with an easy-to-use interface. And second, to expose enough of the adjustable parameters (i.e. genetic operators, and evolutionary configuration) to make the simulation useful for teaching evolutionary algorithms to undergraduate students. Code, Description

Developer, Biolume: Evolution in Action Art Exhibit

URL: http://adamwbrown.net/biolume-header1-jpg/
Description: The Biolume art exhibit is meant to captivate and inform the general public. The installation will comprise approximately 150 culptural robots that ‘evolve’ to better interact with patrons. Through interaction with the public, Biolume robots gain energy and are preferentially selected for reproduction to ‘replace’ less fit neighbors.