Soft Robots Learn to Crawl

Explore the cutting-edge world of soft robotics in this 56-minute conference talk by Audry Sedal from Montreal Robotics. Dive into a comprehensive framework for simulating, co-optimizing, and transferring soft legged robot designs from simulation to reality. Discover how the compliance of soft robots provides "mechanical intelligence," allowing for passive behaviors that are challenging to program traditionally. Learn about the intricate coupling between mechanical design and control in soft robotics, and how co-optimization techniques can generate sophisticated robots. Examine the challenges of simulating soft robot dynamics and the innovative solution of combining finite element simulation with model order reduction techniques. Explore a reinforcement learning-based framework for co-optimization and witness successful examples of optimized, constructed, and zero-shot sim-to-real transferred soft crawling robots. Compare the performance of learned robots to expert-designed crawlers, showcasing the potential for generating novel, high-performing designs in well-understood domains. Gain insights from Audry Sedal, an Assistant Professor at McGill University and Associate Member at Mila, who leads the MACRObotics group focusing on Morphology, Actuation, and Computation for Robotics.