Exploiting Intrinsic Stochasticity of Real-Time Simulation to Facilitate Robust Reinforcement Learning for Robot Manipulation

Simulation is essential to reinforcement learning (RL) before implementation in the real world, especially for safety-critical applications like robot manipulation. Conventionally, RL agents are sensitive to the discrepancies between the simulation and the real world, known as the sim-to-real gap. The application of domain randomization, a technique used to fill this gap, is limited to the imposition of heuristic-randomized models. We investigate the properties of intrinsic stochasticity of real-time simulation (RT-IS) of off-the-shelf simulation software and its potential to improve the robustness of RL methods and the performance of domain randomization. Firstly, we conduct analytical studies to measure the correlation of RT-IS with the occupation of the computer hardware and validate its comparability with the natural stochasticity of a physical robot. Then, we apply the RT-IS feature in the training of an RL agent. The simulation and physical experiment results verify the feasibility and applicability of RT-IS to robust RL agent design for robot manipulation tasks. The RT-IS-powered robust RL agent outperforms conventional RL agents on robots with modeling uncertainties. It requires fewer heuristic randomization and achieves better generalizability than the conventional domain-randomization-powered agents. Our findings provide a new perspective on the sim-to-real problem in practical applications like robot manipulation tasks.