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

arXiv: https://arxiv.org/abs/2605.28527 · HTML: https://mlsanity.com/paper/2605.28527/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 27 May 2026.

## Built on by (most cited first, 6 of 6)

- [The Linear Representation Hypothesis for Vision-Language-Action Models](https://mlsanity.com/paper/2609.30996/) (2026)
- [The Low-Rank Structure of VLA Reinforcement Learning](https://mlsanity.com/paper/2609.34599/) (2026)
- [Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readouts](https://mlsanity.com/paper/2609.34684/) (2026)
- [eRLT: Efficient VLA Reinforcement Learning via Action-Relevant Token Routing](https://mlsanity.com/paper/2610.00913/) (2026)
- [Do VLAs Understand and Adapt to the Objects They Handle, or Simply Replay Learned Behaviors?](https://mlsanity.com/paper/2610.06078/) (2026)
- [RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes](https://mlsanity.com/paper/2610.11480/) (2026)

## Abstract (arXiv)

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.