Flash-WAM: Modality-Aware Distillation for World Action Models

Arman Akbari, Ci Zhang, Arash Akbari, Lin Zhao, Yixiao Chen, Weiwei Chen, Xuan Zhang, Geng Yuan, Yanzhi Wang · N/A · 2026

Summary

Abstract Summary

World-action models (WAMs) jointly generate future video and robot actions through iterative diffusion, achieving strong performance on manipulation benchmarks but requiring tens of denoising steps, a cost that precludes real-time control. Step distillation has emerged as the natural remedy, but off-the-shelf methods break down in the joint video-action setting because video and action streams use different SNR-shifted noise schedules and reach training with substantially different marginal noise distributions, an asymmetry that single-modality distillation methods cannot accommodate. We introduce Flash-WAM, a modality-aware step-distillation framework inspired by consistency distillation that selects the consistency function for each modality to match its noise regime: a linear-gradient-scaling parametrization for the action stream's low-noise regime, paired with a variance-preserving parametrization for the video stream's high-noise regime, grounded in a structural analysis of the consistency-function family that characterizes the achievable gradient scaling under the consistency boundary condition. Instantiated on LingBot-VA, Flash-WAM compresses inference to a single step in each modality. On RoboTwin 2.0, this reduces per-chunk latency from 8.1 seconds to 348 ms on NVIDIA L40S, a 23x speedup that enables real-time inference. Flash-WAM preserves task success on simulation benchmarks (85.5% RoboTwin 2.0, 95.7% LIBERO) and substantially recovers real-world performance (60% average on a Unitree G1 humanoid robot), while naive consistency distillation drops to 24% at the same step budget.

Key Points

Demonstrates humanoid robot control on real hardware
Addresses dexterous manipulation challenges
Uses simulation or synthetic data for training
Uses knowledge distillation to compress models

Flash-WAM: Modality-Aware Distillation for World Action Models

Authors: Arman Akbari, Ci Zhang, Arash Akbari, Lin Zhao, Yixiao Chen, Weiwei Chen, Xuan Zhang, Geng Yuan, Yanzhi Wang

Venue: arXiv preprint | Year: 2026

arXiv: 2606.05254v1

Abstract

World-action models (WAMs) jointly generate future video and robot actions through iterative diffusion, achieving strong performance on manipulation benchmarks but requiring tens of denoising steps, a cost that precludes real-time control. Step distillation has emerged as the natural remedy, but off-the-shelf methods break down in the joint video-action setting because video and action streams use different SNR-shifted noise schedules and reach training with substantially different marginal noise distributions, an asymmetry that single-modality distillation methods cannot accommodate. We introduce \textbf{Flash-WAM}, a modality-aware step-distillation framework inspired by consistency distillation that selects the consistency function for each modality to match its noise regime: a linear-gradient-scaling parametrization for the action stream’s low-noise regime, paired with a variance-preserving parametrization for the video stream’s high-noise regime, grounded in a structural analysis of the consistency-function family that characterizes the achievable gradient scaling under the consistency boundary condition. Instantiated on LingBot-VA, Flash-WAM compresses inference to a single step in each modality. On RoboTwin 2.0, this reduces per-chunk latency from $8.1$ seconds to $348$ ms on NVIDIA L40S, a $23{\times}$ speedup that enables real-time inference. Flash-WAM preserves task success on simulation benchmarks ($85.5%$ RoboTwin 2.0, $95.7%$ LIBERO) and substantially recovers real-world performance ($60%$ average on a Unitree G1 humanoid robot), while naive consistency distillation drops to $24%$ at the same step budget.

Key Contributions

Demonstrates humanoid robot control on real hardware
Addresses dexterous manipulation challenges
Uses simulation or synthetic data for training
Uses knowledge distillation to compress models

Topics

sim-to-real
reinforcement-learning
diffusion-policy
manipulation
humanoid

Code & Data

No code repository linked in paper metadata.

BibTeX

@article{Akbari2026_260605254v1,
  title={Flash-WAM: Modality-Aware Distillation for World Action Models},
  author={Arman Akbari and Ci Zhang and Arash Akbari and Lin Zhao and Yixiao Chen and Weiwei Chen and Xuan Zhang and Geng Yuan and Yanzhi Wang},
  year={2026},
  eprint={2606.05254v1},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2606.05254v1}
}