Abstract: | Evolutionary computing techniques have shown initial promise in automating th design of controllers for simple robotic tasks. However, their application is relatively new to robotics, and their limitations with respect to complex tasks are not yet fully understood. This project studies the prospect of the scalability of evolving robot controllers by applying the technique to a ball-pushing task. Results, though inconclusive, highlight the need for better tools specialised to evolutionary robotics simulation, and for more emphasis on the study of its underlying theory.
|