aiqbench analysis
CAC-VLA introduces a novel framework that integrates latent-action conditioning directly into Vision-Language Models, addressing the critical gap between multimodal understanding and precise motor control in robotics. By training the VLM to predict structured representations rather than executable trajectories, this approach improves success rates on standard benchmarks like LIBERO and CALVIN without requiring separate action-generation frameworks. This advancement represents a significant step forward in making generalist robot manipulation more efficient and robust through native model optimization.
CAC-VLA: Context-Gated Action Conditioning for Vision-Language-Action Models
arXiv:2607.04816v2 Announce Type: replace 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, LIBERO-Plus, and CALVIN demonstrate the effectiveness of CAC-VLA, achieving 98.9% and 90.4% average success rates on LIBERO and LIBERO-Plus, respectively, and outperforming {\pi}0.5 by 9.3 percentage points on CALVIN.
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, LIBERO-Plus, and CALVIN demonstrate the effectiveness of CAC-VLA, achieving 98.9% and 90.4% average success rates on LIBERO and LIBERO-Plus, respectively, and outperforming {\pi}0.5 by 9.3 percentage points on CALVIN.
Robotics (cs.RO)
arXiv:2607.04816 [cs.RO]
(or arXiv:2607.04816v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2607.04816 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 267 words
Related coverage
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · Sep 4, 2026
APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies
arXiv · Jun 11, 2026
RoboICL: Embodied In-Context Learning with GPT-6 Astra
arXiv · Sep 30, 2026
Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore
AWS · Oct 16, 2026