Learning Visual Parkour from Generated Images

TL;DR

This paper introduces a simulation approach for robot visual parkour using generative models to synthesize realistic image sequences, enabling zero-shot transfer to real-world RGB observations with a low-cost camera.

Contribution

It presents a novel method to incorporate realistic visual data into robot simulation, facilitating zero-shot transfer from simulation to real-world RGB perception.

Findings

Successful zero-shot transfer to real-world robot with RGB camera

Generative models produce diverse, physically accurate scene images

Enhanced simulation realism improves robot learning outcomes

Abstract

Fast and accurate physics simulation is an essential component of robot learning, where robots can explore failure scenarios that are difficult to produce in the real world and learn from unlimited on-policy data. Yet, it remains challenging to incorporate RGB-color perception into the sim-to-real pipeline that matches the real world in its richness and realism. In this work, we train a robot dog in simulation for visual parkour. We propose a way to use generative models to synthesize diverse and physically accurate image sequences of the scene from the robot's ego-centric perspective. We present demonstrations of zero-shot transfer to the RGB-only observations of the real world on a robot equipped with a low-cost, off-the-shelf color camera. website visit https://lucidsim.github.io