VLA-OPD: Bridging Offline SFT and Online RL for Vision-Language-Action Models via On-Policy Distillation

• URL: http://arxiv.org/abs/2603.26666v1
• Date: Fri, 27 Mar 2026 17:59:33 GMT
• Title: VLA-OPD: Bridging Offline SFT and Online RL for Vision-Language-Action Models via On-Policy Distillation
• Abstract summary: On-Policy VLA Distillation (VLA-OPD) is a framework bridging the efficiency of online Reinforcement Learning (RL) with the robustness of offline Supervised Fine-Tuning (SFT)
• Score: 14.33516456804297
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Although pre-trained Vision-Language-Action (VLA) models exhibit impressive generalization in robotic manipulation, post-training remains crucial to ensure reliable performance during deployment. However, standard offline Supervised Fine-Tuning (SFT) suffers from distribution shifts and catastrophic forgetting of pre-trained capabilities, while online Reinforcement Learning (RL) struggles with sparse rewards and poor sample efficiency. In this paper, we propose On-Policy VLA Distillation (VLA-OPD), a framework bridging the efficiency of SFT with the robustness of RL. Instead of relying on sparse environmental rewards, VLA-OPD leverages an expert teacher to provide dense, token-level supervision on the student's self-generated trajectories. This enables active error correction on policy-induced states while preserving pre-trained general capabilities through gentle alignment. Crucially, we formulate VLA-OPD via a Reverse-KL objective. Unlike standard Forward-KL that induces mode-covering entropy explosion, or Hard-CE that causes premature entropy collapse, our bounded mode-seeking objective ensures stable policy learning by filtering out the teacher's epistemic uncertainty while maintaining action diversity. Experiments on LIBERO and RoboTwin2.0 benchmarks demonstrate that VLA-OPD significantly improves sample efficiency over RL and robustness over SFT, while effectively mitigating catastrophic forgetting during post-training.

Related papers

• ARMOR: Stabilizing On-Policy LLM RL with Off-Policy Anchor Samples [65.74946647574696]
  Reinforcement learning (RL) has significantly enhanced the reasoning capabilities of large language models (LLMs)<n>In this work, we investigate a critical source of this instability: over-optimization.<n>We propose ARMOR, a framework that shifts the paradigm from passive penalty to active sample stabilization.
  arXiv  Detail & Related papers  (2026-07-11T21:22:46Z)
• ROAD-VLA: Robust Online Adaptation via Self-Distillation for Vision-Language-Action Models [11.1063401404824]
  We propose an advantage-guided self-distillation framework that constructs a teacher directly in action space by perturbing action-token logits with calibrated advantage estimates.<n>Across seven robotic manipulation environments with in-distribution and out-of-distribution shifts, ROADVLA outperforms PPO in nearly all settings.
  arXiv  Detail & Related papers  (2026-06-24T13:17:59Z)
• Jump-Start Reinforcement Learning with Vision-Language-Action Regularization [1.2599533416395767]
  Reinforcement learning (RL) enables high-frequency, closed-loop control for robotic manipulation.<n>Current limitations hinder their direct use in fast and precise manipulation.<n>We propose Vision-Language-Action Jump-Starting (VLAJS) to improve exploration and learning efficiency.
  arXiv  Detail & Related papers  (2026-04-15T11:17:54Z)
• Learning from Mistakes: Post-Training for Driving VLA with Takeover Data [22.44026866655379]
  TakeVLA is a novel VLA post-training framework that overcomes shortcomings through two complementary innovations.<n>First, we introduce pre-takeover language supervision, which allows the VLA to learn from mistakes proactively.<n>Second, we propose Scenario Dreaming, a reinforcement fine-tuning paradigm that operates in reconstruceted takeover scenarios.
  arXiv  Detail & Related papers  (2026-03-16T08:33:48Z)
• Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failures [14.313346858887286]
  Vision-Language-Action (VLA) models for autonomous driving often hit a performance plateau during Reinforcement Learning (RL) optimization.<n>This stagnation arises from exploration capabilities constrained by previous Supervised Fine-Tuning (SFT)<n>We propose Explicit Learning from Failures (ELF-VLA), a framework that augments RL with structured diagnostic feedback.
  arXiv  Detail & Related papers  (2026-03-01T11:41:22Z)
• CRL-VLA: Continual Vision-Language-Action Learning [40.18167835795084]
  Continual Reinforcement Learning is a promising pathway for deploying VLA models in lifelong robotic scenarios.<n>We introduce CRL-VLA, a framework for continual post-training of VLA models with rigorous theoretical bounds.<n>We derive a unified performance bound linking the stability-plasticity trade-off to goal-conditioned advantage magnitude, scaled by policy divergence.
  arXiv  Detail & Related papers  (2026-02-03T12:09:53Z)
• A Step Back: Prefix Importance Ratio Stabilizes Policy Optimization [58.116300485427764]
  Reinforcement learning post-training can elicit reasoning behaviors in large language models.<n> token-level correction often leads to unstable training dynamics when the degree of off-policyness is large.<n>We propose a simple yet effective objective, Minimum Prefix Ratio (MinPRO)
  arXiv  Detail & Related papers  (2026-01-30T08:47:19Z)
• On-the-Fly VLA Adaptation via Test-Time Reinforcement Learning [61.38273866396522]
  Vision-Language-Action models have emerged as a powerful paradigm for general-purpose robot learning.<n>Current methods remain unsuitable for challenging simulated- or physical-world deployments.<n>We introduce a Test-Time Reinforcement Learning for VLAs framework that enables on-the-fly policy adaptation during inference.
  arXiv  Detail & Related papers  (2026-01-11T01:51:30Z)
• Trust-Region Adaptive Policy Optimization [82.09255251747818]
  Post-training methods play an important role in improving large language models' (LLMs) complex reasoning abilities.<n>We introduce TRAPO, a framework that interleavesSupervised Fine-Tuning (SFT) and Reinforcement Learning (RL) within each training instance.<n>Experiments on five mathematical reasoning benchmarks show that TRAPO consistently surpasses standard SFT, RL, and SFT-then-RL pipelines.
  arXiv  Detail & Related papers  (2025-12-19T14:37:07Z)
• COVLM-RL: Critical Object-Oriented Reasoning for Autonomous Driving Using VLM-Guided Reinforcement Learning [55.83415345423854]
  We propose COVLM-RL, a novel end-to-end driving framework that integrates Critical Object-oriented (CO) reasoning with reinforcement learning (RL)<n>Experiments conducted in the CARLA simulator demonstrate that COVLM-RL significantly improves the success rate by 30% in trained driving environments.
  arXiv  Detail & Related papers  (2025-12-10T06:18:16Z)
• Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach [78.4812458793128]
  We propose textbfTACO, a test-time-scaling framework that applies a lightweight pseudo-count estimator as a high-fidelity verifier of action chunks.<n>Our method resembles the classical anti-exploration principle in offline reinforcement learning (RL), and being gradient-free, it incurs significant computational benefits.
  arXiv  Detail & Related papers  (2025-12-02T14:42:54Z)
• RobustVLA: Robustness-Aware Reinforcement Post-Training for Vision-Language-Action Models [33.503927352666096]
  Vision-Language-Action (VLA) models fail to generalize reliably in out-of-distribution deployments.<n>We introduce RobustVLA, a lightweight online RL post-training method designed to explicitly enhance the resilience of VLA models.<n>Our results highlight the importance of robustness-aware RL post-training as a key step toward improving the principled reliability and robustness of VLA models.
  arXiv  Detail & Related papers  (2025-11-03T08:30:48Z)
• Human-in-the-loop Online Rejection Sampling for Robotic Manipulation [55.99788088622936]
  Hi-ORS stabilizes value estimation by filtering out negatively rewarded samples during online fine-tuning.<n>Hi-ORS fine-tunes a pi-base policy to master contact-rich manipulation in just 1.5 hours of real-world training.
  arXiv  Detail & Related papers  (2025-10-30T11:53:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.