Evolution Gym: A Large-Scale Benchmark for Evolving Soft Robots


Both the design and control of a robot play equally important roles in its task performance. However, while optimal control is well studied in the machine learning and robotics community, less attention is placed on finding the optimal robot design. This is mainly because co-optimizing the robot design and control is characterized as a challenging problem, and more importantly, a comprehensive evaluation benchmark for robot co-optimization does not exist. In this paper, we propose Evolution Gym, the first large-scale benchmark for evolving the design of soft robots. In our benchmark, each robot is composed of different types of voxels (e.g., soft, rigid, actuators), resulting in a modular and expressive robot design space. Our benchmark environments span a wide range of tasks, including locomotion on various types of terrains and manipulation. Furthermore, we develop several robot evolution algorithms by combining state-of-the-art design optimization methods and deep reinforcement learning techniques. By evaluating the algorithms on our benchmark platform, the results demonstrate that the performance of robots continuously increases as the evolution progresses. Additionally, even though the robots are evolved autonomously from scratch without prior knowledge, they often grow to resemble existing natural creatures while outperforming hand-designed robots. Nevertheless, all tested algorithms fail to find robots that succeed in our hardest environments. This suggests that more advanced algorithms are required to explore the high dimensional design space and evolve increasingly intelligent robots - an open problem for future research. Our website with code, environments, and videos is available at https://sites.google.com/view/evolution-gym-benchmark

NeurIPS 2021


Evolution Gym is a toolkit for developing and comparing algorithms for co-optimizing design and control of soft robots, covering 30+ tasks spanning locomotion on various types of terrains and manipulation.