Decoding Task Progress from VLA Representations

Atiksh Bhardwaj, Edward Weiyi Duan, Prithwish Dan, Wei-Chiu Ma, Preston Culbertson · N/A · 2026

Summary

Abstract Summary

Vision-language-action models (VLAs) are moving rapidly towards deployment as general-purpose manipulation policies, but we currently lack basic tools for understanding what these models represent internally or for monitoring them at runtime. Leveraging ideas from mechanistic interpretability, we probe the residual stream of $π_{0.5}$ and find that task progress, the normalized time remaining in a trajectory, is linearly readable from the activations. We find that this signal is present in the pretrained PaliGemma backbone prior to training on any robot-specific data. A single linear probe generalizes to unseen tasks and varies under language counterfactuals when trained on multi-prompt data, but does not enable meaningful steering of the policy. These properties make the signal directly useful for instrumenting deployed VLAs. We use the probe as a simple label-free OOD detector, which detects stalled task progress, and find it competitive with state-of-the-art methods. Our results suggest that VLAs have rich, linearly readable internal representations of semantic quantities like task progress, and that learning to read these signals offers a lightweight, interpretable path toward monitoring deployed visuomotor policies.

Key Points

Vision-language-action models (VLAs) are moving rapidly towards deployment as general-purpose man...
Leveraging ideas from mechanistic interpretability, we probe the residual stream of $π_{0.5}$ and...
We find that this signal is present in the pretrained PaliGemma backbone prior to training on any...
A single linear probe generalizes to unseen tasks and varies under language counterfactuals when ...
These properties make the signal directly useful for instrumenting deployed VLAs

Decoding Task Progress from VLA Representations

|Authors: Atiksh Bhardwaj, Edward Weiyi Duan, Prithwish Dan, Wei-Chiu Ma, Preston Culbertson

|Venue: arXiv preprint | Year: 2026

|arXiv: 2608.13474v1

Abstract

Key Contributions

Vision-language-action models (VLAs) are moving rapidly towards deployment as general-purpose man…
Leveraging ideas from mechanistic interpretability, we probe the residual stream of $π_{0.5}$ and…
We find that this signal is present in the pretrained PaliGemma backbone prior to training on any…
A single linear probe generalizes to unseen tasks and varies under language counterfactuals when …
These properties make the signal directly useful for instrumenting deployed VLAs

Topics

manipulation
vision
vla
reinforcement-learning
control

Code & Data

No code repository linked in paper metadata.

BibTeX

@article{Bhardwaj2026_260813474v1,
  title     = {Decoding Task Progress from VLA Representations},
  author    = {Atiksh Bhardwaj and Edward Weiyi Duan and Prithwish Dan and Wei-Chiu Ma and Preston Culbertson},
  year      = {2026},
  eprint    = {2608.13474v1},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2608.13474v1}
}