Offline Reinforcement Learning for Visual Navigation

TL;DR

This paper introduces ReViND, an offline reinforcement learning system for robotic navigation that utilizes existing datasets to optimize user-defined rewards, enabling effective real-world navigation without additional data collection.

Contribution

ReViND is the first offline RL system for robotic navigation that leverages pre-existing data to adapt to user-specified reward functions in real-world scenarios.

Findings

Successfully navigates to distant goals using only offline data

Exhibits behaviors aligned with different user-defined rewards

Operates effectively without additional data collection or fine-tuning

Abstract

Reinforcement learning can enable robots to navigate to distant goals while optimizing user-specified reward functions, including preferences for following lanes, staying on paved paths, or avoiding freshly mowed grass. However, online learning from trial-and-error for real-world robots is logistically challenging, and methods that instead can utilize existing datasets of robotic navigation data could be significantly more scalable and enable broader generalization. In this paper, we present ReViND, the first offline RL system for robotic navigation that can leverage previously collected data to optimize user-specified reward functions in the real-world. We evaluate our system for off-road navigation without any additional data collection or fine-tuning, and show that it can navigate to distant goals using only offline training from this dataset, and exhibit behaviors that qualitatively differ based on the user-specified reward function.