Vision language action model learns when specialists should take control
RouteRLT: Learning When and Which RL Specialist Should Control a Vision-Language-Action Policy
Robotics
Summary
Vision-language-action models can do many tasks but often struggle with tricky parts that need careful control, like plugging in cables. The authors designed RouteRLT, a system that decides when a generalist model should hand control to more specialized controllers that are better at precise actions. Their method learns to switch between these specialists without needing prior knowledge of task stages. Tests show RouteRLT improves performance in both simulations and real robot tasks, keeping the strengths of the general model while improving precision when it matters.
What this means in practice
- •For industrial robot programmers: Automatically route control between generalist and specialist policies for precise robot tasks like cable insertion and object manipulation.
- •For automation integrators: Improve robot task reliability in contact-rich manufacturing by deploying learned routing to manage multiple control policies for different task phases.
Authors
Chongyu Zhu, Jaden Hinds, Hyegang Kim, Juan Sebastian Rojas, Ramy Elmallah, Chi-Guhn Lee
Abstract
Vision-language-action (VLA) models provide broad manipulation competence, but often struggle during the precision-critical stages that dominate contact-rich industrial tasks such as connector insertion and cable management. A common remedy is to refine a pretrained VLA with reinforcement learning (RL), enabling task-specific improvement beyond behavior cloning. However, how to preserve its generalist behavior while deciding when RL refinement is needed and which specialized policy should act remains an open question. In this work, we present RouteRLT, a routing framework that learns when and which RL specialist, an RL policy trained for a single precision-critical phase, should take control from a generalist VLA. A phase selector identifies the active controller, a stabilizer suppresses transient switches, and an action-boundary manager handles transitions between chunked policy outputs. We evaluate RouteRLT on multi-object pick-and-place tasks in LIBERO, as well as on a real-world cable pickup and port-insertion task with multiple precision-critical stages. In simulation, the learned routing improves over the base VLA and matches routing with privileged phase boundaries, without accessing those boundaries at deployment. The real-robot evaluation validates automatic routing to both the pickup and insertion specialists under an operator-aligned handoff protocol. Altogether, these results show that learned routing applies RL specialist control where precise adaptation is most valuable while preserving generalist VLA behavior, including recovery from failed execution attempts.