Guide Your Agent with Adaptive Multimodal Rewards

TL;DR

This paper introduces ARP, a framework that uses multimodal embeddings and natural language instructions to improve agent generalization in unseen environments, effectively addressing goal misgeneralization in imitation learning.

Contribution

The paper proposes a novel adaptive reward mechanism using pre-trained multimodal encoders and fine-tuning to enhance agent generalization with natural language instructions.

Findings

ARP outperforms existing text-conditioned policies in unseen environments.

Multimodal reward signals improve agent adaptation and goal achievement.

Fine-tuning encoders further boosts performance.

Abstract

Developing an agent capable of adapting to unseen environments remains a difficult challenge in imitation learning. This work presents Adaptive Return-conditioned Policy (ARP), an efficient framework designed to enhance the agent's generalization ability using natural language task descriptions and pre-trained multimodal encoders. Our key idea is to calculate a similarity between visual observations and natural language instructions in the pre-trained multimodal embedding space (such as CLIP) and use it as a reward signal. We then train a return-conditioned policy using expert demonstrations labeled with multimodal rewards. Because the multimodal rewards provide adaptive signals at each timestep, our ARP effectively mitigates the goal misgeneralization. This results in superior generalization performances even when faced with unseen text instructions, compared to existing text-conditioned policies. To improve the quality of rewards, we also introduce a fine-tuning method for pre-trained multimodal encoders, further enhancing the performance. Video demonstrations and source code are available on the project website: \url{https://sites.google.com/view/2023arp}.