Driving · 2026-10-15

Atomic-WAM: Execution-Grounded Atomic-Motion Reasoning for World Action Models

Shijun Shi, Zhe Liu, Zhi Hou, Yinan Chen, Tao Zhang, Kai Hu, Yong Liu, Jiangning Zhang

arXiv:2610.20157PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

Atomic-WAM is an understand-then-execute world action model for robot control. A vision-language reasoner converts task instructions plus visual context into state-dependent atomic-motion descriptions, which condition a WAM to predict future video and an action chunk. The authors build temporally aligned, contact-enriched atomic-motion supervision from multi-view DROID demonstrations and post-train the reasoner with execution-grounded GRPO (EG-GRPO).

Why it matters

The authors address two gaps in instruction-to-execution: semantic grounding (translating language into concrete interactions) and temporal granularity (selecting the next local motion as the scene evolves). By using atomic-motion descriptions and then adapting the reasoner from execution outcomes, Atomic-WAM targets closed-loop compatibility between high-level motion selection and the downstream executor.

Method

  • Understand-then-execute interface: a VLM reasoner predicts the next state-dependent atomic-motion description from task instruction and visual context; a WAM predicts future video latents and an action chunk conditioned on the description and current observation/state.
  • Atomic-motion data curation: automatic pipeline partitions multi-view DROID demonstrations into contact-enriched, temporally aligned atomic-motion segments; labels are verified with a checker and include visible grasp/contact details for grasp-related segments.
  • Execution-grounded reasoner adaptation: with the WAM fixed, the reasoner is post-trained using EG-GRPO, where rewards depend on motion appropriateness (task reward) and execution success of the frozen WAM’s rollouts.

Limitation

During EG-GRPO training, computing the execution reward requires assessing every local rollout, making the process costly in inference latency and token usage; the authors use human judgments of execution success and note future work on an efficient automatic execution evaluator.

Abstract (from arXiv)

World Action Models (WAMs) typically adapt video foundation models to robotic control through joint video and action prediction. However, task-level instructions may leave the next physical interaction underspecified, requiring semantic grounding and state-dependent motion selection. To this end, we introduce Atomic-WAM, an understand-then-execute framework that connects task reasoning to action generation through atomic-motion descriptions. Specifically, a vision-language reasoner describes the next local motion from the task instruction and visual context, guiding a WAM to predict future video and an action chunk. Furthermore, an automatic annotation pipeline curates temporally aligned, contact-enriched supervision from multi-view DROID demonstrations. To make atomic-motion predictions executable by the WAM, we post-train the reasoner with Execution-Grounded GRPO (EG-GRPO). With the WAM fixed, EG-GRPO adapts the reasoner through rewards for motion appropriateness and execution success. Finally, Atomic-WAM achieves 39.6% task success under default instructions and 31.6% under vague instructions in a real-to-sim evaluation on RoboLab. Project page: https://ssj9596.github.io/Atomic-WAM-project

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-14 · Indexed by Pub-AI: 2026-10-15 UTC

preprint

Atomic-WAM: Execution-Grounded Atomic-Motion Reasoning for World Action Models

Shijun Shi, Zhe Liu, Zhi Hou, Yinan Chen, Tao Zhang, Kai Hu, Yong Liu, Jiangning Zhang

Preprint posted: 2026-10-14 · Original record ↗

arxiv
2610.20157
Explicit source versions (1)
  • v1 · 2026-10-14
  1. Preprint available

    Source date: 2026-10-14 · First detected: 2026-10-25 UTC · Historical baseline

    Source evidence (1)
Abstract · arxiv

World Action Models (WAMs) typically adapt video foundation models to robotic control through joint video and action prediction. However, task-level instructions may leave the next physical interaction underspecified, requiring semantic grounding and state-dependent motion selection. To this end, we introduce Atomic-WAM, an understand-then-execute framework that connects task reasoning to action generation through atomic-motion descriptions. Specifically, a vision-language reasoner describes the next local motion from the task instruction and visual context, guiding a WAM to predict future video and an action chunk. Furthermore, an automatic annotation pipeline curates temporally aligned, contact-enriched supervision from multi-view DROID demonstrations. To make atomic-motion predictions executable by the WAM, we post-train the reasoner with Execution-Grounded GRPO (EG-GRPO). With the WAM fixed, EG-GRPO adapts the reasoner through rewards for motion appropriateness and execution success. Finally, Atomic-WAM achieves 39.6% task success under default instructions and 31.6% under vague instructions in a real-to-sim evaluation on RoboLab. Project page: https://ssj9596.github.io/Atomic-WAM-project