Data-driven Policy Transfer with Imprecise Perception Simulation

TL;DR

This paper introduces a pipeline for learning continuous robot motion control policies using a non-differentiable physics simulator, combined with a generative model to refine simulation outputs, enabling effective policy transfer to real robots.

Contribution

It proposes a coarse-to-fine learning approach that jointly optimizes motion policies and a generative model for improved transfer from simulation to real-world robots.

Findings

Successful transfer of policies to real robot platform

Enhanced simulation accuracy through generative model refinement

Effective policy learning with non-differentiable physics simulator

Abstract

The paper presents a complete pipeline for learning continuous motion control policies for a mobile robot when only a non-differentiable physics simulator of robot-terrain interactions is available. The multi-modal state estimation of the robot is also complex and difficult to simulate, so we simultaneously learn a generative model which refines simulator outputs. We propose a coarse-to-fine learning paradigm, where the coarse motion planning is alternated with imitation learning and policy transfer to the real robot. The policy is jointly optimized with the generative model. We evaluate the method on a real-world platform in a batch of experiments.