Robotics, Evolution, and Art Lab- Department of Computer Science – IT University of Copenhagen

We work on designing robotic morphologies and controllers for locomotion tasks by using evolutionary algorithms. Specifically, we employ modular robots, simple and autonomous devices that can be assembled together to achieve a functional robot. In ABACUS2.0, we run physical simulations to evaluate the suitability of a robotic morphology and a controller. As these simulations are very time consuming, they are carried out in parallel in a cluster. Our results show that certain module´s geometries and generative encodings can greatly speed up the optimization process. Additionally, the best robots can be assembled and tested in reality, which give us a measurement of their performance. Comparing real and simulated results, we can shed light on how to reduce the reality gap.