Physics-Based Simulation of Agile Robotic Systems

Abstract Development and testing of industrial robot environments is hampered by the limited availability of hardware resources. Simulations provide a more accessible and readily modifiable alternative to physical testing, but also require careful design to maximize their fidelity to the real system. In this paper, we describe new progress in building an entirely physics-driven simulation of the Agility Performance of Robotic Systems (APRS) Laboratory at the National Institute of Standards and Technology (NIST). To maximize the accuracy of physics-based interactions in this environment, we develop several general-purpose improvements to the simulation of parallel grippers and multi-object collisions. We use pick-and-place tasks from the APRS Laboratory as well as generic benchmarks to verify the performance of our simulation. We demonstrate that our proposed improvements result in more physically-consistent simulations compared to standard implementations, regardless of the choice of physics engine or simulation parameters.