Towards the targeted environment-specific evolution of robot components

This research considers the task of evolving the physical structure of a robot to enhance its performance in various environments, which we cast as a significant problem in the field of Evolutionary Robotics. Borrowing from the fields of evolutionary art and sculpture, we evolve only unactuated components of a robot, which simplifies the optimisation problem compared to traditional approaches that must simultaneously evolve both (actuated) body and brain, whilst focusing fidelity onto areas of the robot most likely to benefit from shape optimisation. Our approach uses a Genetic Algorithm to optimise collections of Bezier splines that together create a hexapod leg, with leg performance evaluated in a high-fidelity simulator. The leg is represented in the simulator as a file that can be readily 3D-printed. Experiments are carried out in three different environments; results show different leg shapes are automatically discovered depending on the environment. This proof-of-concept represents an important step towards the environment-dependent optimisation of performance-critical components of standard, capable robots that can carry out a wide variety of tasks.