CAC-VLA: Context-Gated Action Conditioning for Vision-Language-Action Models

2026-07-06

AI Summary

Existing vision-language-action models lack explicit optimization for action-conditioned generation, leading to a disconnect between semantic understanding and precise motor control. This work proposes a context-gated action-conditioning framework that integrates a lightweight latent action interface within the vision-language model itself. By employing a coarse-to-fine latent action representation, the framework adaptively guides action experts to generate continuous control signals. Notably, this approach introduces a native action-conditioning mechanism into vision-language models for the first time, leveraging context gating to dynamically modulate the influence of latent actions on policy decisions—without requiring any additional planning modules. Evaluated on the LIBERO and LIBERO-Plus benchmarks, the method achieves average success rates of 98.3% and 89.5%, respectively, substantially advancing performance in continuous action control.

Abstract

Vision-Language-Action (VLA) models have become a promising paradigm for generalist robot manipulation, where visual-language representations are used to condition continuous action generation. However, these representations are not explicitly optimized for action conditioning, leaving the action expert to bridge the gap between multimodal understanding and precise motor control. Recent action-reasoning methods introduce additional modules to generate explicit action plans or action-space reasoning signals, demonstrating the benefit of action-level guidance but often requiring separate action-generation frameworks. We propose CAC-VLA, a Context-Gated Action Conditioning framework that learns a lightweight latent-action interface directly within the VLM. Instead of generating executable trajectories, CAC-VLA trains the VLM to predict coarse-to-fine latent actions, which are structured representations encoded from future action segments, and adaptively leverages them to condition the action expert via a context gate. This enables VLM-native action conditioning while calibrating the influence of latent-action guidance on expert action generation. Experiments on LIBERO and LIBERO-Plus demonstrate the effectiveness of CAC-VLA, achieving 98.3% average success rate on LIBERO and 89.5% LIBERO-Plus, suggesting that context-gated latent-action conditioning is an effective interface for continuous expert control.

Problem

• Vision-Language-Action models
• action conditioning
• latent actions
• robot manipulation
• multimodal understanding

Innovation

• latent-action conditioning
• context gating
• vision-language-action models
• action generation
• robot manipulation