Driving · 2026-10-14
Box-WAM: Box Trajectories as Compact State Representations for World Action Models
Qinxi Yu, Wenzhou Ding, Gloria X. Wang, Albert J. Zhai, Shenlong Wang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
BoxWAM replaces dense future video prediction in world action models with oriented 3D bounding box (OBB) trajectories of task-relevant objects. The model jointly denoises future box tokens and action tokens in a rectified-flow transformer with causal state-action attention. The authors report lower future-modeling cost and improved action prediction, evaluated in simulation (RoboTwin2.0) and on a real robot.
Why it matters
The paper addresses what compact future-state representation preserves the planning/supervision benefits of future prediction while reducing cost. By predicting interpretable object-centric box trajectories instead of full frames, BoxWAM aims to provide a lightweight intermediate representation that helps action generation, with evidence in both simulation and real-world robotic manipulation.
Method
- BoxWAM predicts future object-centric state as image-plane coordinates of the eight corners of an oriented 3D bounding box (16-D per timestep) for actively manipulated objects.
- A rectified-flow transformer jointly denoises future box tokens and action tokens; an enforced causal attention lets action tokens attend to future state tokens while blocking future state tokens from attending to future action tokens.
- At training, BoxWAM uses a rectified-flow objective over the joint future variable of box trajectories and actions; at inference it integrates the learned ODE with Euler integration to produce action chunks for execution.
Limitation
The paper does not report quantitative reconstruction-error evaluation for the real-world oriented bounding box (OBB) supervision reconstruction pipeline.
Abstract (from arXiv)
World action models (WAMs) are a category of policy models that ex- plicitly predict both actions and future world states. They have been shown to outperform action-only models across various robotic manipulation tasks. How- ever, representing the future with full-frame video prediction can be computation- ally expensive during both training and inference. This raises a central question: what compact state representation preserves the planning and supervision bene- fits of future prediction? In this paper, we propose BoxWAM, which replaces dense future frames with oriented bounding box trajectories of task-relevant ob- jects. By explicitly imagining object-centric future motion, our method provides a lightweight prediction target that captures the key physical changes induced by robot actions. We show that box trajectory prediction substantially reduces com- putational cost while improving action prediction performance, suggesting that object-centric compact states can serve as an efficient alternative to video-based world action modeling for robotic manipulation. Result videos can be found on https://box-wam.github.io
Related papers
- TrajWAM: An Efficient Trajectory-Guided World Action Model for Dynamic Robotic Manipulation
- {\Delta}WAM: Distilling Action Tangent Fields into World Action Models
- WAVE-0: Bridging VLA Policies and WAMs with Lightweight Latent Future Prediction
- Copper-Policy: Focus on the Representation for Robust Robot Manipulation
- Self-Distilling Future Video Representations for World Action Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-13 · Indexed by Pub-AI: 2026-10-14 UTC
Box-WAM: Box Trajectories as Compact State Representations for World Action Models
Preprint posted: 2026-10-13 · Original record ↗
- arxiv
- 2610.17526
Explicit source versions (1)
- v1 · 2026-10-13
- Preprint available
Source date: 2026-10-13 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World action models (WAMs) are a category of policy models that ex- plicitly predict both actions and future world states. They have been shown to outperform action-only models across various robotic manipulation tasks. How- ever, representing the future with full-frame video prediction can be computation- ally expensive during both training and inference. This raises a central question: what compact state representation preserves the planning and supervision bene- fits of future prediction? In this paper, we propose BoxWAM, which replaces dense future frames with oriented bounding box trajectories of task-relevant ob- jects. By explicitly imagining object-centric future motion, our method provides a lightweight prediction target that captures the key physical changes induced by robot actions. We show that box trajectory prediction substantially reduces com- putational cost while improving action prediction performance, suggesting that object-centric compact states can serve as an efficient alternative to video-based world action modeling for robotic manipulation. Result videos can be found on https://box-wam.github.io