Unified Visuomotor Targets: Supervising VLAs Beyond Physical Actions

2026-08-04

AI Summary

This work addresses the semantic gap between high-level visual-language representations and low-level motor actions in existing Vision-Language-Action (VLA) models, which hinders learning efficiency and generalization. To bridge this gap without altering model architecture or requiring additional data, the paper proposes Unified Visual-Tactile Targets (UVT), a framework that jointly encodes motor control signals and visual scene dynamics into a unified latent prediction target. By reframing supervision as a reconstruction objective, UVT aligns perceptual understanding with action execution at a semantic level. Experiments demonstrate that UVT substantially improves training efficiency, task success rates, and policy robustness across multiple simulated and real-world bimanual manipulation tasks, with particularly pronounced gains under data-scarce conditions and in complex environments.

Abstract

VLA models are trained to predict robot actions from visual and language observations. This is a natural choice, but it creates a mismatch: VLMs encode rich, high-level representations of scenes and goals, while robot actions are low-level signals with limited task structure. We ask whether changing what the policy is trained to predict, rather than how it is architecturally designed, can yield better and more efficiently trained policies. We propose UVT (Unified Visuomotor Target), a unified latent prediction target that jointly encodes motor control and visual scene transition information, requiring no architectural changes and no additional data. Applied to two representative VLA systems across simulation benchmarks and real bimanual manipulation tasks, UVT improves training efficiency, final task performance, and policy robustness, with particularly strong gains under limited training budgets and challenging environmental conditions. Rollout videos and additional qualitative results are available at our project webpage: https://unified-visuomotor-targets.github.io/

Problem

• Visuomotor Learning
• Visual Language Models
• Robot Action Prediction
• Representation Mismatch
• Policy Training

Innovation

• Unified Visuomotor Target
• Visual-Language-Action Models
• Latent Prediction Target
• Policy Training Efficiency
• Robotic Manipulation