The Dark Side of Rich Rewards: Understanding and Mitigating Noise in VLM Rewards

TL;DR

This paper investigates the negative impact of false positive rewards in Vision-Language Model-based reward signals for embodied agents, introduces BiMI to reduce noise, and demonstrates improved learning efficiency.

Contribution

It identifies false positive rewards as particularly harmful, analyzes the limitations of cosine similarity, and proposes BiMI as a novel reward function to mitigate reward noise.

Findings

False positive rewards significantly hinder learning.

BiMI improves training efficiency in navigation tasks.

Cosine similarity is prone to false positive errors.

Abstract

While Vision-Language Models (VLMs) are increasingly used to generate reward signals for training embodied agents to follow instructions, our research reveals that agents guided by VLM rewards often underperform compared to those employing only intrinsic (exploration-driven) rewards, contradicting expectations set by recent work. We hypothesize that false positive rewards -- instances where unintended trajectories are incorrectly rewarded -- are more detrimental than false negatives. Our analysis confirms this hypothesis, revealing that the widely used cosine similarity metric is prone to false positive reward estimates. To address this, we introduce BiMI ({Bi}nary {M}utual {I}nformation), a novel reward function designed to mitigate noise. BiMI significantly enhances learning efficiency across diverse and challenging embodied navigation environments. Our findings offer a nuanced understanding of how different types of reward noise impact agent learning and highlight the importance of addressing multimodal reward signal noise when training embodied agents