ElasticFlow: One-Step Physics-Consistent Policy with Elastic Time Horizons for Language-Guided Manipulation

TL;DR

ElasticFlow is a novel physics-consistent, one-step diffusion policy framework that enables efficient, high-frequency robotic control aligned with semantic instructions, outperforming existing methods on long-horizon tasks.

Contribution

We introduce ElasticFlow, a one-step, physics-consistent diffusion policy with Elastic Time Horizons, addressing latency and physical consistency issues in embodied AI control.

Findings

Achieves approximately 71Hz inference speed.

Outperforms state-of-the-art methods on long-horizon tasks.

Demonstrates effective alignment of control granularity with semantic instructions.

Abstract

Diffusion policies have demonstrated exceptional performance in embodied AI. However, their iterative denoising process results in high latency, and existing acceleration methods often sacrifice physical consistency. To address this, we propose ElasticFlow, a distillation-free, physics-consistent one-step policy framework. We reconstruct the Mean Field Theory by directly modeling the average velocity field, enabling a direct single-step mapping from noise to action. Addressing the Temporal Heterogeneity of robotic tasks, we introduce the Elastic Time Horizons mechanism. This mechanism effectively overcomes Spectral Bias by explicitly encoding control granularity, achieving efficient alignment between semantic instructions and physical execution horizons. Experiments on benchmarks such as LIBERO, CALVIN, and RoboTwin demonstrate that ElasticFlow achieves efficient 1-NFE inference (approximately 71Hz). Furthermore, it outperforms state-of-the-art methods, including OpenVLA and $\pi_0$, on long-horizon tasks, highlighting its potential for efficient, robust, and semantically aligned control.