StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching

Key points

• 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.1
• Existing methods typically address this issue by either improving inference efficiency or enhancing action continuity across prediction chunks, without jointly addressing both challenges.1
• 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.1
• 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.1

Sources (1)

• [1]StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 12:56 PM

  To bridge this gap, we propose StreamingWAM, a streaming-diffusion framework for WAM-based robot control.

  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.

Extractive summary: sentences quoted from the sources.