Learning Fused State Representations for Control from Multi-View Observations

TL;DR

This paper introduces MFSC, a multi-view fusion method for reinforcement learning that learns compact, task-relevant state representations, improving robustness and performance even with missing or noisy views.

Contribution

The paper proposes MFSC, integrating bisimulation metric learning and auxiliary tasks to enhance multi-view state representation learning in control tasks.

Findings

MFSC outperforms existing methods in multi-view RL tasks.

MFSC maintains high performance with missing or noisy views.

The auxiliary mask and reconstruction tasks improve robustness.

Abstract

Multi-View Reinforcement Learning (MVRL) seeks to provide agents with multi-view observations, enabling them to perceive environment with greater effectiveness and precision. Recent advancements in MVRL focus on extracting latent representations from multiview observations and leveraging them in control tasks. However, it is not straightforward to learn compact and task-relevant representations, particularly in the presence of redundancy, distracting information, or missing views. In this paper, we propose Multi-view Fusion State for Control (MFSC), firstly incorporating bisimulation metric learning into MVRL to learn task-relevant representations. Furthermore, we propose a multiview-based mask and latent reconstruction auxiliary task that exploits shared information across views and improves MFSC's robustness in missing views by introducing a mask token. Extensive experimental results demonstrate that our method outperforms existing approaches in MVRL tasks. Even in more realistic scenarios with interference or missing views, MFSC consistently maintains high performance.