Vision language action models speed up robot tasks with smart timing
PaceVLA: Adaptive Temporal Action Quantization in Vision-Language-Action Models
RoboticsArtificial Intelligence
Summary
Robots often move slowly because they repeat many small steps exactly as shown in training, even when some steps could be combined safely. The authors created a method called PaceVLA that lets robots decide when to act with simple, faster commands and when to use detailed, careful moves. This way, robots finish tasks more quickly without making more mistakes, working well in both virtual tests and real-world situations.
What this means in practice
- •For robotics engineers: Develop robot control systems that automatically pick between detailed and coarse actions for faster task completion without losing accuracy.
- •For manufacturing automation teams: Deploy adaptable robot policies that speed up manipulation tasks on assembly lines while maintaining precision in sensitive steps.$Commercial implications: Enables more efficient industrial robots that reduce cycle times and increase throughput in automated production settings.
Authors
Hojin Jeon, Jimin Lee, Taekwan Kim, Nayoung Oh, Taeyoung Kim, Myungkyu Koo, Jinmo Koo, Joonwoo Ahn, Jinwoo Shin
Abstract
Fast and reliable manipulation is crucial for practical robot deployment. Although Vision-Language-Action (VLA) models have recently shown remarkable progress as generalist robot policies, their execution often remains temporally inefficient. A key source of this inefficiency is that policies learn fine-grained temporal patterns from teleoperated demonstrations, requiring many low-level action steps. Temporal action quantization offers a simple way to accelerate execution by aggregating consecutive low-level actions into temporally coarser actions. However, applying action quantization without considering when fine-grained control is required introduces a speed-accuracy trade-off: accelerated actions may efficiently reduce redundant steps in some phases, while discarding critical motion details in precision-sensitive interactions. To address this, we propose PaceVLA, a unified framework for adaptive temporal execution in VLAs that integrates two coupled components: (1) mode-specific action decoders built on a shared action expert, supporting multiple temporal granularities and prediction horizons, and (2) a quantizability-guided mode selection scheme which decides whether coarse execution is appropriate, and a context-conditioned router that selects an execution mode at each inference step. Experiments across three multi-task simulation benchmarks and six real-world tasks show that PaceVLA improves execution efficiency while maintaining task success rate across multiple embodiments, including long-horizon and dexterous manipulation.