Agile Soft Robot Arms

Control policies for soft robot arms typically assume quasi-static motion or require a hand-designed motion plan. In this work, we demonstrate real-time motion planning and control of agile maneuvers by soft robot arms. We use reinforcement learning and key insights with simulation and actuator modeling to overcome sim-to-real challenges, enabling zero-shot sim-to-real transfer. Shown below is an example of our framework applied to a shelf-reaching task, in which we want the tip of the soft robot arm to reach a target location on a shelf. I presented this work with my co-first author at the 2023 Conference on Robot Learning. Check out the project website here.