analiza aiqbench
TORL-VLA addresses a critical gap in Vision-Language-Action models by enabling online adaptation for contact-rich robotic tasks where static offline policies fail under distribution shifts. The framework combines a tactile-derived wrench-aware VLA with a lightweight reinforcement learning module to refine actions in real-time, while an intervention-censored critic stabilizes training on mixed human and policy data. Real-robot evaluations demonstrate improved success rates and execution efficiency for complex manipulation scenarios like egg handling and latch mechanisms.
TORL-VLA: Tactile Guided Online Reinforcement Learning for Contact-Rich Manipulation
arXiv:2606.09337v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models have become a powerful framework for robotic manipulation, and recent studies have introduced tactile or force feedback into VLAs to address contact-rich tasks. However, these models are typically deployed as offline policies. When contact conditions shift from the training distribution, the policy cannot perform online adaptation, leading to problems such as inappropriate contact forces and inefficient retries. Therefore, we propose TORL-VLA, a tactile-guided online reinforcement learning framework that couples tactile feedback with policy refinement for contact-rich manipulation. Our method introduces a tactile-derived wrench-aware VLA to predict reference actions and future wrench sequences, while a lightweight online RL module is used to refine the reference actions. To stabilize learning from mixed exploratory policy-generated and human-intervention data, we introduce an intervention-censored critic that prevents post-intervention success from being wrongly credited to policy-generated actions preceding intervention. Real-robot experiments on long-horizon contact-rich tasks, including latch manipulation, coffee-cup placement, and egg handling, show that TORL-VLA improves success rates at both subtask and full-task levels, as well as time-bounded execution efficiency over strong baselines.
Abstract:Vision-Language-Action (VLA) models have become a powerful framework for robotic manipulation, and recent studies have introduced tactile or force feedback into VLAs to address contact-rich tasks. However, these models are typically deployed as offline policies. When contact conditions shift from the training distribution, the policy cannot perform online adaptation, leading to problems such as inappropriate contact forces and inefficient retries. Therefore, we propose TORL-VLA, a tactile-guided online reinforcement learning framework that couples tactile feedback with policy refinement for contact-rich manipulation. Our method introduces a tactile-derived wrench-aware VLA to predict reference actions and future wrench sequences, while a lightweight online RL module is used to refine the reference actions. To stabilize learning from mixed exploratory policy-generated and human-intervention data, we introduce an intervention-censored critic that prevents post-intervention success from being wrongly credited to policy-generated actions preceding intervention. Real-robot experiments on long-horizon contact-rich tasks, including latch manipulation, coffee-cup placement, and egg handling, show that TORL-VLA improves success rates at both subtask and full-task levels, as well as time-bounded execution efficiency over strong baselines.
Robotics (cs.RO)
arXiv:2606.09337 [cs.RO]
(or arXiv:2606.09337v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.09337 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Text integral sub licența deschisă a publicației · 264 cuvinte
Subiecte conexe
TORL-VLA: Tactile Guided Online Reinforcement Learning for Contact-Rich Manipulation
arXiv · 11 iun. 2026
TORL-VLA: Tactile Guided Online Reinforcement Learning for Contact-Rich Manipulation
arXiv · 17 iun. 2026
ViTaR: Visuo-Tactile Residual Adaptation for Foundation VLA Manipulation
arXiv · 18 aug. 2026
TacCoRL: Integrating Tactile Feedback into VLA via Simulation
arXiv · 12 iun. 2026