Fastopd cuts robot vision model latency while keeping strong performance
FastOPD: On-Policy Distillation for Lightweight VLA Deployment
RoboticsArtificial IntelligenceComputer Vision and Pattern RecognitionMachine Learning
Summary
Vision-language-action models help robots understand and interact with their surroundings, but large models can be too slow for real-world use. The authors propose FastOPD, a technique to create smaller, faster models by learning from bigger ones without losing much ability. They show that their approach keeps much of the original model's skill while speeding up decision time significantly. This makes it easier to deploy advanced robot control on real hardware.
What this means in practice
- •For robotics engineers: Deploy lightweight robot control models that maintain strong performance but run much faster on real hardware.
- •For embedded ai developers: Build compact vision-language-action systems suitable for devices with limited computing power using efficient distillation.
Authors
Yoojin Oh, Jeongsol Kim, Yeonwoo Seo, Jangho Park, Seonghyun Jin, Sunwoo Park, Youngmin Kim, Youngjun Jun, Kyumin Choi, Jong Chul Ye
Abstract
Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging. Existing approaches typically mitigate this issue by designing smaller architectures or reducing the iterative denoising steps in flow-based policies. In this work, we propose FastOPD, a foundation-to-lightweight VLA framework that enables the practical deployment of large-scale VLAs through efficient on-policy distillation. Specifically, FastOPD adapts a flow map for single-state teacher supervision and combines it with a self-consistency objective to construct a compact student that learns the teacher dynamics. Furthermore, we theoretically demonstrate that minimizing this objective allows the distilled student to recover a distribution on par with that induced by an ideal few-step teacher model. We evaluate FastOPD across diverse foundation policies in simulation and real-world experiments. On LIBERO, FastOPD retains 84% of the performance of $π_{0.5}$ with only two inference steps, reducing inference latency by 78.1% while outperforming existing few-step distillation baselines in average success rate. With LingBot-VLA as the teacher, FastOPD improves the single-step success rate over the base student by 15.9 percentage points on RoboTwin 2.0. We further demonstrate its applicability to a World Action Model (WAM) and deploy a compact student distilled from MolmoAct2 on a real robot.