DROPO: Sim-to-Real Transfer with Offline Domain Randomization

TL;DR

DROPO is a new offline method for estimating domain randomization distributions that improves sim-to-real transfer in robotic manipulation by explicitly modeling parameter uncertainty from limited data.

Contribution

DROPO introduces a likelihood-based approach to estimate domain randomization distributions using only offline data, enhancing safe sim-to-real transfer.

Findings

Successfully recovers dynamic parameter distributions in simulation.

Achieves better zero-shot transfer performance than prior methods.

Handles unmodeled phenomena effectively.

Abstract

In recent years, domain randomization over dynamics parameters has gained a lot of traction as a method for sim-to-real transfer of reinforcement learning policies in robotic manipulation; however, finding optimal randomization distributions can be difficult. In this paper, we introduce DROPO, a novel method for estimating domain randomization distributions for safe sim-to-real transfer. Unlike prior work, DROPO only requires a limited, precollected offline dataset of trajectories, and explicitly models parameter uncertainty to match real data using a likelihood-based approach. We demonstrate that DROPO is capable of recovering dynamic parameter distributions in simulation and finding a distribution capable of compensating for an unmodeled phenomenon. We also evaluate the method in two zero-shot sim-to-real transfer scenarios, showing successful domain transfer and improved performance over prior methods.