Titulo Estágio
Evolving robot’s brains and bodies with genetic programming and grammatical evolution
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI-FCTUC
Enquadramento
Robots come in various forms and sizes. While they have been applied in the industry for quite some time, they are becoming increasingly popular throughout society, taking on part of the daily chores. However, the process of designing a robot for performing a particular task is not straightforward, often requiring considerable effort from trained experts. Moreover, there is also the risk that, once deployed, the robot will underperform and even be unable to cope with unforeseen circumstances. As an example, consider the problem of developing a robot to explore the lunar surface. While it is possible to build simulations of the environment to aid in the development of such a robot, it is still possible that mismatches between the real environment and simulation, or even unforeseen conditions, render the robot unable to succeed in its task. A better approach would be to send a module capable of building robots autonomously, along with an algorithm that provided the designs for those robots. That is the ultimate goal of Evolutionary Robotics (ER), a field of research that draws inspiration from nature to devise algorithms that evolve robots as the biological species did. While the existing literature on ER dates back over 20 years, there are still many challenges in evolving both the bodies and brains of the robots, with most works focusing only on the latter.
Objetivo
The main objective of this dissertation is to evolve the bodies and brains of robots for a given task. The evolution will be carried out in simulators such as PyBullet or Webots. The student should review the literature and choose the most suitable methods for evolving the body and the brain. However, a viable option is to employ grammar-based methods for evolving the body. In turn, Artificial Neural Networks (ANN) may be used as the brains, effectively creating end-to-end controllers that receive the inputs from the robot’s sensors and output motor commands. These ANNs may be completely evolved using one of the existing Neuroevolution approaches (e.g., NEAT [1]) or by an Evolutionary Reinforcement Learning approach.
[1] - Stanley, K. O., & Miikkulainen, R. (2002). Evolving neural networks through augmenting topologies. Evolutionary computation, 10(2), 99-127.
Plano de Trabalhos - Semestre 1
- Literature review.
- Implementation and test of the most promising approaches from the literature.
- Development of an evolutionary approach to evolve the controller (brain) of a hand-designed robot.
- Writing of the intermediate report.
Plano de Trabalhos - Semestre 2
- Development of the evolutionary approach to co-evolve the robot’s bodies and brains.
- Validation of the proposed methods and comparison with the existing approaches.
- Writing of the dissertation.
- Writing of a scientific article with the main results.
Condições
The work shall be carried out within CISUC’s Bio-Inspired Artificial Intelligence group (bAI), under the supervision of Prof. João Macedo. Additionally, eligible students may have the opportunity to receive a scholarship (Bolsa de Investigação para Licenciado) following the monthly stipend guidelines set by Fundação para a Ciência e Tecnologia (FCT).
Orientador
João Macedo
jmacedo@dei.uc.pt 📩