Robotics and Evolutionary Computing
Google Scholar
You can also find out about my published research on Google Scholar, which, as well as providing links to the papers and their abstracts, additionally provides the latest citation counts and index values.
Overview
While I was working on my PhD I attended a workshop in Cairns given by the famous Rodney Brooks introducing his work on behaviour-based robotics. I was immediately fascinated and after completing my PhD I set about purchasing some of my own robots and experimenting with Brooks’ ideas. These robots became the RoboCoasters, a robot soccer team that represented Griffith University in the FIRA robot soccer world cups in Rockhampton, Beijing and Seoul in the early 2000s. The team reached the quarter-finals in Seoul where they were defeated by the eventual champions from Germany.
The picture on the right shows our first three robots. They would chase an orange golf ball across a ping-pong table watched over by a giant Sony video camera (things were not very sophiticated in those days).
My main work on these robots was the development of a robust vision system to recognise the shapes situated on top of the robots (see the picture above and the FIRA 2002 paper). However, my most enjoyable moment was the success of my implementation of one of Braitenberg’s simple vehicle algorithms to control each robot’s ball chasing behaviour. Before doing this I had written pages of code to control the robot motors based on mathematical projections of where I wanted each robot to be in the next few seconds. In constrast, my implementation of Braitenberg’s algorithm used about 4 lines of code. All it did was to make the amount of power sent to each of the two wheel motors proportional to the angle between the direction the robot was facing and the position of the ball. So if the ball were 45 degrees to the right, the left wheel would get power proportional to x+45 and the right wheel would get power proportional to x-45, where x is the current desired speed of the robot. This would make the robot turn slightly toward the ball in that particular time frame (bearing in mind this command would be rerun 50 times per second). Once the robot was pointing toward the ball, the angle would be zero, and each wheel would get the same power, so it would go straight ahead. The only other condition was that x itself was inversely proportional to the absolute value of the ball’s angle (i.e. the closer the angle gets to zero, regardless of whether the ball is to the left or right, the greater is x and the faster the robot moves). Given these simple rules the robot immediately started to smoothly and effectively turn and chase the ball almost as if it had a mind of its own.
I also branched out from my focus on pure constraint satisfaction by experimenting with evolutionary algorithms, even though these algorithms were still intended to generate new constraint satisfaction algorithms (e.g. see CEC 2004). This approach emerged from an experience I had with a local search algorithm that was outperforming all my other prototypes, and that I thought was working according to my very clever design. But when I was writing up a paper describing the algorithm, I found I had made a coding error and the only reason the algorithm was doing so well was because of that error. I now came to see that most of the successful refinements I had made to the algorithms I had developed were primarily a matter of trial and error and not a matter of deep insight. So while the core ideas – like that of constraint weighting, or of Braitenberg’s vehicle control algorithm – were a matter of insight, the details of the implementation that produced the best results could not be foreseen. Hence, I came to think it would be better to automate this process of trial and error rather than exhaustively try out all the possible algorithms by hand. This idea was taken up by one of my PhD students, Stuart Bain, who then developed a genetic programming approach to generate and test new constraint satisfaction algorithms, which produced the four papers below and culminated in the publication of his PhD.
2005
Bain, S., Thornton, J. R. & Sattar, A. (2005). Evolving Variable-Ordering Heuristics for Constrained Optimisation. CP 2005: 11th International Conference on Contraint Programming. Lecture Notes in Computer Science, 3709, 732-736, Springer, ISSN 0302-9743.
Abstract: In this paper we present and evaluate an evolutionary approach for learning new constraint satisfaction algorithms, specifically for MAX-SAT optimisation problems. Our approach offers two signicant advantages over existing methods: it allows the evolution of more complex combinations of heuristics, and; it can identify fruitful synergies among heuristics. Using four different classes of MAX-SAT problems, we experimentally demonstrate that algorithms evolved with this method exhibit superior performance in comparison to general purpose methods.
Bain, S., Thornton, J. R. & Sattar, A. (2005). A Comparison of Evolutionary Methods for the Discovery of Local Search Heuristics. AI 2005: 18th Australian Joint Conference on Artificial Intelligence. Lecture Notes in Computer Science, 3809, 1068-1074, Springer, ISSN 0302-9743.
Abstract: Methods of adaptive constraint satisfaction have recently become of interest to overcome the limitations imposed on “black-box” search algorithms by the no free lunch theorems. Two methods that each use an evolutionary algorithm to adapt to particular classes of problem are the CLASS system of Fukunaga and the evolutionary constraint algorithm work of Bain et al. We directly compare these methods, demonstrating that although special purpose methods can learn excellent algorithms, on average standard evolutionary operators perform even better, and are less susceptible to the problems of bloat and redundancy.
2004
Bain, S., Thornton, J. R. & Sattar, A. (2004). Methods of Automatic Algorithm Generation. PRICAI 2004: 8th Pacific Rim International Conference on Artificial Intelligence. Lecture Notes in Computer Science, 3157, 144-153, Springer, ISSN 0302-9743.
Abstract: Many methods have been proposed to automatically generate algorithms for solving constraint satisfaction problems. The aim of these methods has been to overcome the difficulties associated with matching algorithms to specific constraint satisfaction problems. This paper examines three methods of generating algorithms: a randomised search, a beam search and an evolutionary method. The evolutionary method is shown to have considerably more flexibility than existing alternatives, being able to discover entirely new heuristics and to exploit synergies between heuristics.
Bain, S., Thornton, J. R. & Sattar, A. (2004). Evolving Algorithms for Constraint Satisfaction. In: CEC 2004: Proceedings of the 2004 Congress on Evolutionary Computation, IEEE, pp. 265-272.
Abstract: This paper proposes a framework for automatically evolving constraint satisfaction algorithms using genetic programming. The aim is to overcome the difculties associated with matching algorithms to specic constraint satisfaction problems. A representation is introduced that is suitable for genetic programming and that can handle both complete and local search heuristics. In addition, the representation is shown to have considerably more exibility than existing alternatives, being able to discover entirely new heuristics and to exploit synergies between heuristics. In a preliminary empirical study, it is shown that the new framework is capable of evolving algorithms for solving the well-studied problem of boolean satisability testing.
2003
Leonard, J., Treffner, P., & Thornton, J. R. (2003). Tau Guidance for Mobile Soccer Robots. Studies in Perception and Action VII, 169-172, Psychology Press, ISBN 978-0805848052.
Abstract: Traditional approaches to mobile robot guidance have utilised an internal model of the environment constructed from sensor data in order to plan a course of action. Although this approach has been challenged by behaviour-based robotics (e.g., Brooks, 1991), the creation of “smart perceptual instruments” that attempt to directly couple perception and action has been seriously addressed by the ecological robotics paradigm founded upon Gibson’s ecological optics (Duchon et al., 1995). We explore the possibility of using tau information for guidance of a mobile robot. More specifically, an attempt to implement some of the navigation techniques using tau as outlined by Lee (1998) was undertaken. Problems encountered during implementation and possible solutions are given. Although Lee introduced the concept of using tau (τ) for the guidance of movement (Lee, 1998) and although tau is usually said to represent the time to closure of a gap, this is not entirely true. A negative tau represents the time to closure of the gap, while a positive tau represents the time since the closure of the gap (i.e., the gap is opening). A gap can represent a distance, an angle, a difference in force or any other gap. The information on time to closure provided by a sensor can therefore be used to control the closure of a gap and thus to guide movement. Lee also introduced the idea of guidance through tau coupling in which the ratio of taus of different gaps are kept invariant.
2002
Ferreira Jr., V., Thornton, J. R. & Leonard, J. (2002). A Subsumption Architecture for Robotic Soccer. In: Proceedings of the 2002 FIRA Robot World Congress, KAIST (Korea Advanced Institute of Science and Technology), pp. 648-654.
Abstract: The purpose of this paper is to report in our investigation about the feasibility of implementing a pure subsumption architecture for controlling our team of Microsot Robotic Soccer players, the RoboCoasters. This extends previous work on subsumption by considering a highly dynamic and competitive environment where speed and accuracy are paramount. The major finding of our study is a rigid adherence to the incremental development concept of subsumption has proved impractical when applied to our single architecture approach to robotic soccer. This finding has led us to propose two extensions to subsumption that simplify the development process and address the need for continual improvement in lower level behaviours. These are: (a) The creation of a chief behaviour as a permanent top layer for the architecture; and (b) the development of modules that can be changed within a layer without changing the overall layer’s structure.
Thornton, J. R., Leonard, J., Wiseby, R. & Lee, Y. (2002). Shape Recognition and Enhanced Control Systems for Robot Soccer. In: Proceedings of the 2002 FIRA Robot World Congress, KAIST (Korea Advanced Institute of Science and Technology), pp. 670-675.
Abstract: In this paper, we report on the current developments of our Mirosot Robot Soccer team, the RoboCoasters, in preparation for the FIRA Robot World Cup 2002 in Korea. We describe a number of enhancements to both the vision systems and the control system. Vision enhancements include a unique shape recognition system, expanded from recognition of three shapes to five shapes for the middle league competition. Control enhancements include improvements to our goalkeeper and attacker’s behaviour, and the introduction of role swapping between the robots. Finally, we outline our proposed future research for the team.