Real-World Reinforcement Learning of Active Perception Behaviors

TL;DR

This paper introduces AAWR, a simple and efficient method for training active perception policies in robots, leveraging privileged sensors during training to improve performance under partial observability.

Contribution

The paper presents AAWR, a novel training approach that uses privileged sensors and demonstrations to quickly learn active perception behaviors in robots.

Findings

AAWR outperforms prior methods on 8 manipulation tasks across 3 robots.

It enables robots to operate effectively under severe partial observability.

AAWR efficiently generates information-gathering behaviors from suboptimal initial policies.

Abstract

A robot's instantaneous sensory observations do not always reveal task-relevant state information. Under such partial observability, optimal behavior typically involves explicitly acting to gain the missing information. Today's standard robot learning techniques struggle to produce such active perception behaviors. We propose a simple real-world robot learning recipe to efficiently train active perception policies. Our approach, asymmetric advantage weighted regression (AAWR), exploits access to "privileged" extra sensors at training time. The privileged sensors enable training high-quality privileged value functions that aid in estimating the advantage of the target policy. Bootstrapping from a small number of potentially suboptimal demonstrations and an easy-to-obtain coarse policy initialization, AAWR quickly acquires active perception behaviors and boosts task performance. In evaluations on 8 manipulation tasks on 3 robots spanning varying degrees of partial observability, AAWR synthesizes reliable active perception behaviors that outperform all prior approaches. When initialized with a "generalist" robot policy that struggles with active perception tasks, AAWR efficiently generates information-gathering behaviors that allow it to operate under severe partial observability for manipulation tasks. Website: https://penn-pal-lab.github.io/aawr/