European Space Agency - Ariadna Project:
Evolution in Robotic Islands
Official Study Description
Evolutionary Robotics is a technique which has recently received growing attention from the robotics research community, as it promises the automatic synthesis of robot controllers. Such an automated framework is based on the use of artificial evolution to reinforce learning within robot populations, by effectively tuning the parameters of randomly generated controllers, where the controllers are typically artificial neural networks (ANNs).
The parallelisation of artificial evolution has been extensively studied in the context of global optimisation, and has been shown to significant speedup of the optimisation process. In addition, island-based simulations have demonstrated that the migration of individuals between independent evolutionary runs improves the performance of the optimisation process, both in terms of function evaluations required, and in terms of the quality of the solutions obtained, providing a better balance between exploration and exploitation of the search space. This technique has also been applied to high dimensional and difficult engineering problems.
The island evolution paradigm has potential in the design of automated space robots for various reasons:
- It could significantly speedup the design process by exploiting parallelism, while improving the quality of the solutions found, relative to traditional evolutionary techniques.
- It could relieve the experimenter from a significant part of the burden required to properly set up the evolutionary process. For example, poorly-tuned algorithms cooperating via migration have been shown to work as well as a single instance of a well-tuned algorithm in global optimisation.
- When designing controllers for robots, the exchange of individuals between islands, corresponds to the introduction of new solutions. This increases the diversity in a population, and the via recombination of genetic material might endow agents with capabilities that would be extremely unlikely to evolve in a single run; in other words, it could facilitate the progressive composition of a rich behavioural repertoire.
The main objective of this study is to use island evolution to optimise the neuro-controller of a model Mars rover, with a vision to empirically demonstrate an improvement in automated design.
Postscript
This study was completed in 2010. Results can be found in the listed publications.
Publications
- Peniak M., Bentley B., Marocco D., Cangelosi A., Ampatzis C.,
Izzo D. & Biscani F. (2010) An Evolutionary Approach to
Designing Autonomous Planetary Rovers. Proceedings of TAROS2010 Towards Autonomous Robotic Systems, August 31 - September 2, 2010, Plymouth, UK. (pdf)
- Peniak M., Bentley B.,
Marocco D., Cangelosi A., Ampatzis C., Izzo D. & Biscani F. (2010)
An Island-Model Framework for Evolving Neuro-Controllers for Planetary
Rover Control. Proceedings of IJCNN2010 International Joint Conference on Neural Networks, July 18 - 23, 2010, Barcelona, Spain. (pdf)
-
Cangelosi A., Marocco D., Peniak M., Bentley B., Ampatzis C. & Izzo D. (2010) Evolution
in robotic islands - optimising the design of autonomous robot
controllers for navigation and exploration of unknown environments. Advanced Concepts Team, The European Space Agency; Noordwijk, The Netherlands. Ariadna-ID: 09-8301 (pdf)
- Peniak M., Bentley B., Marocco D., Ampatzis C., Biscani F., Izzo
D. & Cangelosi A. (2010) Designing Autonomous Robot Controllers for
Planetary Exploration: A Model of a Mars Rover. Proceedings of PCCAT2010 Postgraduate Conference for Computing: Applications and Theory, June 9, 2010, Exeter, UK. (pdf) - Awarded Best Paper
Project Poster
Ariadna project meeting at the
European Space Research and Technology Centre (ESTEC) - July 2010
|
|
