Chrono: A multi-physics engine for simulation of robophysical systems

Early-stage computer analysis of robophysical systems informs critical design decisions, where systematic study of the design-space via experiment can be challenging and expensive. We present the latest capabilities of a multi-physics platform, called Chrono, that allows for computer modeling and simulation of such systems. The term multi-physics is used herein as a broad umbrella for rigid body dynamics with frictional-contact, flexible body dynamics, fluid dynamics, and fluid-solid interaction. This paradigm is critical in many robophysical systems since locomotion of rigid/flexible robots in conjunction with rigid, granular, and fluid substrates is ubiquitous. The simulation platform supports applications that range from locomotion of robotic agents over non-trivial geometries and on gravel, to underwater robotics. Furthermore, we describe how these simulations can be used as episodes of deep reinforcement learning models for training smart agents who can learn efficient locomotion pathways through machine learning. This is important because (i) fine-tuning the optimum design parameters can be laborious, and (ii) training agents in a virtual environment instead of the real world reduces the time and risks of the operation.