Learning to Act Robustly with View-Invariant Latent Actions

TL;DR

This paper introduces VILA, a method for learning view-invariant, physically grounded latent actions to improve the robustness of vision-based robotic policies against viewpoint changes, enhancing generalization and transferability.

Contribution

VILA models latent actions based on physical dynamics and aligns them across viewpoints, providing a novel approach for view-invariant policy learning in robotics.

Findings

VILA enables policies to generalize to unseen viewpoints.

VILA improves transfer to new tasks in real-world experiments.

VILA enhances robustness of vision-based robotic policies.

Abstract

Vision-based robotic policies often struggle with even minor viewpoint changes, underscoring the need for view-invariant visual representations. This challenge becomes more pronounced in real-world settings, where viewpoint variability is unavoidable and can significantly disrupt policy performance. Existing methods typically learn invariance from multi-view observations at the scene level, but such approaches rely on visual appearance and fail to incorporate the physical dynamics essential for robust generalization. We propose View-Invariant Latent Action (VILA), which models a latent action capturing transition patterns across trajectories to learn view-invariant representations grounded in physical dynamics. VILA aligns these latent actions across viewpoints using an action-guided objective based on ground-truth action sequences. Experiments in both simulation and the real world show that VILA-based policies generalize effectively to unseen viewpoints and transfer well to new tasks, establishing VILA as a strong pretraining framework that improves robustness and downstream learning performance.