What Frozen VLAs Already Know About Success: A Probing Study of Value-Like Structure in Foundation Robot Policies

• URL: http://arxiv.org/abs/2605.28527v1
• Date: Wed, 27 May 2026 14:23:29 GMT
• Title: What Frozen VLAs Already Know About Success: A Probing Study of Value-Like Structure in Foundation Robot Policies
• Abstract summary: Vision-language--action (VLA) policies are trained to imitate actions; their loss never asks them to estimate reward, progress, or future success.<n>We recover Monte-Carlo outcome targets using lightweight linear probes on frozen features.<n>The gains are not universal and require additional inference compute, but the underlying finding is clean.
• Score: 36.91260665881213
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Vision--language--action (VLA) policies are trained to imitate actions; their loss never asks them to estimate reward, progress, or future success. Their frozen representations nevertheless carry such information, and it can be read out and used to guide action choice without retraining the policy. From mixed successful and failed manipulation trajectories on LIBERO-Goal, we recover Monte-Carlo outcome targets using lightweight linear probes on frozen features. The targets are consistently predictable from OpenVLA, Pi0.5, DINOv2, and CLIP features, and substantially less so from baselines built on progress, time-to-go, task identity, or proprioception. To rule out task and temporal shortcuts, we evaluate the probes under same-task, same-timestep matched comparisons: Pi0.5 probes still reach roughly 92% pairwise ordering accuracy, while label-shuffled controls stay at chance. Used as a test-time selector over sampled Pi0.5 action prefixes, the same probe turns this offline finding into behavior: on push-plate, success rises from 26.7% under greedy decoding to 44.3%, with a second positive case on wine-rack. The gains are not universal and require additional inference compute, but the underlying finding is clean: frozen VLAs already encode information about success that their imitation objective never explicitly demands.

Related papers

• ReDraft, Don't Just Distill: Reference-Driven Revision for Continual VLLM Post-Training [66.72708893922605]
  Post-training of large multimodal models should add new capabilities while preserving those from pre-training.<n>SFT gives explicit target supervision that learns a task from near-zero accuracy, but its off-policy targets move the model far enough to cause forgetting.<n>We introduce ReDraft, which obtains both from the model's own failures.
  arXiv  Detail & Related papers  (2026-09-15T04:59:03Z)
• On-Policy Self-Distillation in Diffusion Models [89.32656931795293]
  Reinforcement learning can align diffusion models with human preferences and task-specific objectives, but endpoint rewards do not specify how an intermediate denoising prediction should change.<n>We introduce DiffusionOPSD as an on-policy self-distillation framework that converts image-level reward guidance into explicit targets for clean-output predictions at sampled queries.<n>Across SD 3.5-M and the step-distilled Z-Image-Turbo, our approach achieves the best final held-out scores in 19 of 20 reward-matched settings across two backbones and ten evaluators.
  arXiv  Detail & Related papers  (2026-08-25T14:55:24Z)
• Reuse Before You Retrieve: Diagnosing Headroom and Complementarity for Test-Time Augmentation of Embodied Multimodal Policies [56.86456077356948]
  Frozen-language-action (VLA) policies are increasingly improved at test time.<n>Additional sampling is useful only when better behavior already exists within the policy's rollouts.<n>Retrieval is most useful when the relevant action prior is not reliably represented by the policy.
  arXiv  Detail & Related papers  (2026-08-18T08:11:31Z)
• Suppression Sticks, Locality Is Fragile: A Closed-Loop Target-and-Control Audit of Task-Vector Negation in VLA Policies [33.432377306905735]
  Task-vector arithmetic offers a closed-form way to modify a model, yet its behavioral locality remains unclear in closed-loop robot control.<n>We present a target-and-control audit of per-skill task-vector subtraction from multitask vision-language-action (VLA) policies.
  arXiv  Detail & Related papers  (2026-08-05T10:59:53Z)
• Continue or Replan? Bernoulli-Continuation Policy Learning for Adaptive Horizon Execution [38.8233224696416]
  Bernoulli-Continuation Policy (BCP) is a lightweight, plug-and-play framework for adaptive horizon execution.<n>On RoboTwin 2.0 with LingBot-VLA as the base policy, BCP improves the average success rate by +11.08% on 13 low-success tasks.<n>On a real robot, BCP lifts success from 74% to 92% and from 44% to 84% on two manipulation tasks.
  arXiv  Detail & Related papers  (2026-08-04T11:21:08Z)
• Look Before You Leap: Distilling Tree Search into Action Evaluation for Frozen VLA Models [60.17417174639865]
  Vision-Language-Action (VLA) models acquire broad embodied capabilities through large-scale pretraining.<n>The prevailing remedy, post-training, improves task-specific performance but narrows the generalist capability that makes pretraining valuable.<n>We propose SVA, a framework that equips frozen VLA policies with long-term consequence awareness.
  arXiv  Detail & Related papers  (2026-07-04T07:53:10Z)
• RouterVLA: Turning Smoke Tests into Supervision for Heterogeneous VLA Selection [5.188057876794867]
  We study whether pre-deployment evaluation rollouts can be reused to supervise policy selection.<n>We find that pre-deployment evaluation rollouts can be reused to supervise policy selection.
  arXiv  Detail & Related papers  (2026-06-25T17:56:33Z)
• When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agents [0.0]
  Long-answer LLM agents can fail quietly: they settle on one reading of the evidence early, then spend the rest of the run defending it.<n>We define representational commitment as cross-run hidden-state convergence at a fixed reasoning step.<n>A runtime monitor detects inconsistent trajectories from hidden states at AUROC up to 0.97 (0.85--0.88 under a stricter split)
  arXiv  Detail & Related papers  (2026-06-22T07:13:13Z)
• Process Rewards with Learned Reliability [25.86885718567082]
  Downstream methods must treat imperfect step-level reward predictions as reliable decision signals.<n>We propose BetaPRM, a distributional PRM that predicts both a step-level success probability and the reliability of that prediction.<n>We show that BetaPRM improves PRM-guided Best-of-N selection while preserving standard step-level error detection.
  arXiv  Detail & Related papers  (2026-05-15T01:57:11Z)
• Done, But Not Sure: Disentangling World Completion from Self-Termination in Embodied Agents [11.233308795768465]
  VIGIL is an evaluation framework that makes terminal commitment independently measurable.<n>Under VIGIL's default protocol, agents observe only egocentric RGB, receive no action-success signals, and must end each episode with a semantic report checked deterministically against hidden world state.<n>This yields two separate scores: world-state completion (W) and benchmark success (B)
  arXiv  Detail & Related papers  (2026-05-09T07:24:07Z)
• Global Prior Meets Local Consistency: Dual-Memory Augmented Vision-Language-Action Model for Efficient Robotic Manipulation [95.89924101984566]
  We introduce OptimusVLA, a dual-memory VLA framework with Global Prior Memory (GPM) and Local Consistency Memory (LCM)<n>GPM replaces Gaussian noise with task-level priors retrieved from semantically similar trajectories.<n>LCM injects a learned consistency constraint that enforces temporal coherence and smoothness of trajectory.
  arXiv  Detail & Related papers  (2026-02-22T15:39:34Z)
• Scaling Verification Can Be More Effective than Scaling Policy Learning for Vision-Language-Action Alignment [58.93227458806748]
  CoVer-VLA is a hierarchical test-time verification pipeline using the trained verifier.<n>Our framework precomputes a diverse set of rephrased instructions from a Vision-Language-Model.<n>It repeatedly generates action candidates for each instruction, and then uses the verifier to select the optimal high-level prompt and low-level action chunks.
  arXiv  Detail & Related papers  (2026-02-12T18:59:59Z)
• Value Vision-Language-Action Planning & Search [1.631000263754549]
  Vision-Language-Action (VLA) models have emerged as powerful generalist policies for robotic manipulation.<n>We introduce Value Vision-Language-Action Planning and Search (V-VLAPS), a framework that augments Monte Carlo Tree Search with a lightweight, learnable value function.<n>We evaluate V-VLAPS on the LIBERO robotic manipulation suite, demonstrating that our value-guided search improves success rates by over 5 percentage points.
  arXiv  Detail & Related papers  (2026-01-02T19:40:34Z)
• Bidirectional Decoding: Improving Action Chunking via Guided Test-Time Sampling [51.38330727868982]
  We show how action chunking impacts the divergence between a learner and a demonstrator.<n>We propose Bidirectional Decoding (BID), a test-time inference algorithm that bridges action chunking with closed-loop adaptation.<n>Our method boosts the performance of two state-of-the-art generative policies across seven simulation benchmarks and two real-world tasks.
  arXiv  Detail & Related papers  (2024-08-30T15:39:34Z)

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.