AI & ML
Stochastic Self-Distillation for Flow-Based Vision-Language-Action Models
arXiv cs.RO0 views
arXiv:2611.20460v1 Announce Type: new Abstract: Flow-based vision-language-action (VLA) models achieve strong performance in robotics through imitation learning, yet fine-tuning them with on-policy reinforcement learning (RL) is limited. In group-based RL methods like Group Relative Policy Optimization (GRPO), every rollout step requires multi-step ODE integration, and the sampler's limited stochasticity leads to rollout groups with low variance. We propose VLA-SSD, which addresses both high inference latency and limited stochasticity. In particular, our method self-distills a pretrained flow-matching action head into a stochastic one-step policy by conditioning a single-jump generator on a Brownian path. We derive a conditional action density via our stochastic one-step policy, enabling efficient post-training via likelihood-ratio RL algorithms. We show that distilling VLA-SSD from the $\pi_0$ VLA reduces the action head's forward-pass time by a factor of 7.7, at the cost of a 5.5% relative decrease in the success rate on the LIBERO benchmark. However, by leveraging the increased sample diversity of VLA-SSD during group generation, post-training a $\pi_{0.5}$ VLA using GRPO yields a 20.6% relative improvement.
Source:https://arxiv.org/abs/2611.20460PreviousXMimic and XSoma: Learning Humanoid Motion Tracking and Locomotion in Under a MinuteNextReRead: Wrong Predictions, Useful Representations - Correcting Frozen Robot Policies by Re-reading Their Own Representations
Related posts
Nvidia Earnings; Office Convergence; OS Wars, AI Edition
0Stratechery
France Is Blaming Iran for the Fridge Cyberattack
0Schneier on Security
Apple iPhone Duo Gets a New Facial Unlock Option With iOS 27.3 Update
0FOSS Bytes
CMM. The cursed language that never really took off.
0dev.to Watercooler

I built a GitHub Action that scores pull requests and never auto-closes them
0dev.to GitHub
I built a Windows app to track Instagram follower and profile changes over time
0dev.to GitHub