SimGAN: Hybrid Simulator Identification for Domain Adaptation via Adversarial Reinforcement Learning
This addresses domain adaptation for robot controllers, particularly in sim-to-real transfer, but is incremental as it builds on existing methods with a hybrid approach.
The paper tackles the problem of transferring learned robot policies to new domains with different dynamics by introducing SimGAN, a framework that identifies a hybrid physics simulator to match simulated trajectories to target domain ones, and shows it outperforms baselines on six robotic locomotion tasks.
As learning-based approaches progress towards automating robot controllers design, transferring learned policies to new domains with different dynamics (e.g. sim-to-real transfer) still demands manual effort. This paper introduces SimGAN, a framework to tackle domain adaptation by identifying a hybrid physics simulator to match the simulated trajectories to the ones from the target domain, using a learned discriminative loss to address the limitations associated with manual loss design. Our hybrid simulator combines neural networks and traditional physics simulation to balance expressiveness and generalizability, and alleviates the need for a carefully selected parameter set in System ID. Once the hybrid simulator is identified via adversarial reinforcement learning, it can be used to refine policies for the target domain, without the need to interleave data collection and policy refinement. We show that our approach outperforms multiple strong baselines on six robotic locomotion tasks for domain adaptation.