EvoScene-VLA: Evolving Scene Beliefs Inside the Action Decoder for Chunked Robot Control

TL;DR

EvoScene-VLA introduces a persistent, action-updated scene state within chunked robot control policies, enhancing multi-step robot task success by integrating recent actions and visual data.

Contribution

It proposes a recurrent scene representation that maintains geometry-aware scene priors across control chunks, improving robot control performance.

Findings

Increases average success rate from 87.2% to 89.1% on RoboTwin tasks.

Outperforms all baselines on the Galaxea R1-Lite robot.

Effectively integrates scene updates with visual observations during control.

Abstract

Chunked vision-language-action (VLA) policies predict multi-step robot controls, conditioning each update on the current visual observation alone. Yet robot actions cause contact, occlusion, and object motion, and the geometry that later decisions depend on can change before the next visual update arrives. Spatial VLAs improve current-frame geometry. Temporal VLAs aggregate past frames. Neither maintains an action-updated scene prior across chunks. We argue for a persistent action-updated scene state across control calls, and introduce EvoScene-VLA. Its recurrent scene prefix carries a geometry-aware scene state across chunks. At each vision-language model (VLM) call, the VLM combines scene information from the current observation with the action-updated prior from the previous chunk; the action decoder outputs both the next action chunk and a compact scene update. This update becomes the next prior, which the VLM corrects against the new observation when the next call arrives. Each control call therefore starts from a scene prior that reflects both recent actions and fresh visual evidence. During training, \textbf{Scene Predictor} supplies future scene-token targets, and Geometric Anchor aligns scene slots with frozen depth and 3D teachers. We discard both modules at deployment. On 31 RoboTwin tasks, EvoScene-VLA raises average success from 87.2% to 89.1% in fixed evaluation and from 86.1% to 88.5% in randomized evaluation. On the Galaxea R1-Lite real robot, EvoScene-VLA outperforms all baselines.