Robotics · 2026-10-14
StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching
Shaojie Ren, Jingwen Sun, Zezhi Liu, Hongjin Chen, Xin Jin, Wei Zou, Zhibo Chen
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
StreamingWAM is a streaming-diffusion world action model for robot control that aims to reduce jerky/discontinuous closed-loop execution. It uses a noise ladder for chunk-wise streaming action prediction, a prefix rollout module for cross-chunk continuity, and dense video-history memory for temporally consistent conditioning, producing competitive results with lower steady-state replanning latency.
Why it matters
The authors target real-world WAM deployment where regenerated action chunks can be costly and can produce abrupt transitions at replanning boundaries. StreamingWAM’s persistent plan buffer with one-step steady-state replanning is designed to improve inference efficiency and action continuity together, and the paper reports latency reductions on a real robot alongside smoother action transitions.
Method
- Noise ladder for streaming: staggers future action blocks across noise levels in a persistent buffer, enabling low-latency single-step steady-state replanning.
- Prefix rollout for continuity: conditions future-action denoising on a clean prefix of recently executed actions to reduce cross-chunk discontinuities.
- Dense video-history memory: uses recent visual observations (o_{t-m+1:t}) as conditioning for temporally consistent action generation.
Limitation
The authors state that using 10-step dynamic deployments for StreamingWAM’s baselines yielded near-zero success because the policy could not react within the task time scale.
Abstract (from arXiv)
World Action Models (WAMs), often built upon pretrained video-generation world models, have emerged as a promising framework for general-purpose and open-ended robot control. However, their real-world deployment often suffers from discontinuous and jerky execution. Existing methods typically address this issue by either improving inference efficiency or enhancing action continuity across prediction chunks, without jointly addressing both challenges. To bridge this gap, we propose StreamingWAM, a streaming-diffusion framework for WAM-based robot control. StreamingWAM combines a noise ladder for chunk-wise action prediction, a prefix rollout module for cross-chunk continuity, and dense video-history memory for temporally consistent conditioning. Experiments on LIBERO, LIBERO-Plus, RoboTwin 2.0, and real-world robotic tasks demonstrate that StreamingWAM improves action continuity across static and dynamic settings, achieving competitive performance while reducing steady-state replanning latency to 72.9 ms in simulation and 57.3 ms on the real robot. The method also produces smoother action transitions, while the streaming staircase reduces sensitivity to the execution horizon K.
Related papers
- Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks
- SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations
- De-Diffusing WAMs: Action-Sufficient Direct Prediction for Robot Control
- WAM-Cache: Staleness-Bounded KV Reuse for Efficient World Action Models
- Long-WAM: Scaling the Context of World-Action Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-11 · Indexed by Pub-AI: 2026-10-14 UTC
StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching
Preprint posted: 2026-10-11 · Original record ↗
- arxiv
- 2610.15585
Explicit source versions (1)
- v1 · 2026-10-11
- Preprint available
Source date: 2026-10-11 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World Action Models (WAMs), often built upon pretrained video-generation world models, have emerged as a promising framework for general-purpose and open-ended robot control. However, their real-world deployment often suffers from discontinuous and jerky execution. Existing methods typically address this issue by either improving inference efficiency or enhancing action continuity across prediction chunks, without jointly addressing both challenges. To bridge this gap, we propose StreamingWAM, a streaming-diffusion framework for WAM-based robot control. StreamingWAM combines a noise ladder for chunk-wise action prediction, a prefix rollout module for cross-chunk continuity, and dense video-history memory for temporally consistent conditioning. Experiments on LIBERO, LIBERO-Plus, RoboTwin 2.0, and real-world robotic tasks demonstrate that StreamingWAM improves action continuity across static and dynamic settings, achieving competitive performance while reducing steady-state replanning latency to 72.9 ms in simulation and 57.3 ms on the real robot. The method also produces smoother action transitions, while the streaming staircase reduces sensitivity to the execution horizon K.