Coarse-to-Control: Action-Token Planning for Vision-Language-Action Models
For robotic manipulation tasks requiring long-horizon planning, this method addresses the compounding error problem by integrating planning directly into the action-token space, offering a practical improvement over existing VLA models.
Coarse-to-Control introduces action-token planning into VLA models, improving long-horizon task performance by first predicting coarse action tokens as a plan before generating executable actions. On LIBERO, SimplerEnv-WidowX, and real-world tasks, it consistently outperforms direct action generation, with largest gains on multi-stage tasks.
Most vision-language-action (VLA) models map observations directly to actions without explicit intermediate planning, which limits performance on long-horizon tasks where early mistakes compound. We propose Coarse-to-Control, a plan-execute VLA that introduces planning natively in the action-token space. The key idea is to let the policy first predict a compact sequence of coarse action tokens that summarize the intended future trajectory, and then generate executable action tokens conditioned on this plan. Because both planning and execution share a unified discrete action vocabulary, the plan stays close to the control manifold and provides directly actionable guidance rather than an abstract hint that must be translated back to motor commands. Experiments on LIBERO, SimplerEnv-WidowX, and real-world manipulation tasks show that action-token planning consistently improves over direct action generation, with the largest gains on long-horizon multi-stage tasks.