Soft Material Modeling and Manipulation

Learning control strategies to manipulate soft materials (e.g., cables, papers) is challenging and the behaviors of soft materials under contact are difficult to model. Collaborating with MIT-IBM Watson AI Lab, we presented a fully differentiable simulation platform for thin-shell object manipulation tasks (e.g., papers) and a differentiable simulation environment for vision-based tactile sensors (e.g., GelSight) for learning robotic skills efficiently in simulation.