Elastica: A Compliant Mechanics Environment for Soft Robotic Control

Soft robots are notoriously hard to control. This is partly due the scarcity of models and simulators able capture their complex continuum mechanics, resulting in a lack control methodologies that take full advantage body compliance. Currently available methods either too computational demanding or overly simplistic physical assumptions, leading paucity simulation resources for developing such schemes. To address this, we introduce Elastica, an open-source environment modeling dynamics soft, slender rods can bend, twist, shear, stretch. We couple Elastica with five state-of-the-art reinforcement learning (RL) algorithms (TRPO, PPO, DDPG, TD3, SAC). successfully demonstrate distributed, dynamic soft robotic arm four scenarios both large action spaces, where RL difficult, small actor must learn interact its environment. Training converges 10 million policy evaluations near real-time evaluation learned policies.