Optical Tactile Sim-to-Real Policy Transfer via Real-to-Sim Tactile Image Translation


Simulation has recently taken a key role in deep reinforcement learning to safely and efficiently generate large datasets for acquiring general and complex control policies. The majority of this research focuses on visual and proprioceptive data, with tactile information often overlooked despite its direct relation to environment interaction. In this work, we present a suite of simulated environments tailored towards tactile robotics and reinforcement learning, including edge/surface exploration and object rolling/balancing tasks. A simple and fast method of simulating optical tactile sensors is provided, where high resolution contact geometry is represented as depth images. Proximal Policy Optimisation (PPO) is used to learn successful policies across all considered tasks. A data-driven approach enables translation of the current state of a real tactile sensor, the TacTip, to corresponding simulated depth images. This policy is implemented within a real-time control loop on a physical robot to demonstrate zero-shot sim-to-real policy transfer over a range of physically-interactive tasks requiring a sense of touch.