Skip to content
Preprint

ODEAction: From Predicting Chunks to Evolving Actions

Qi-Shen Wang Pengxiang Ding Shi-Feng Bao Ming-Yang Sun Heng-Tao Li Yi Gao Bao-Hua Yin Yang Liu Jia-Jun Fu Jia-Cheng Liu Chang-Jing He Pei-Yang Li Run-Ze Suo Yan Wang Xiao-Bo Wang Yong-Hao Dang Cheng Chi
Nov 2026 · 0 citations · 51 references
Computer Science

Abstract

Vision-language-action (VLA) models typically generate actions in chunks, organizing continuous robot execution into separately predicted segments and thus introducing potential discontinuities at action-chunk transitions. We propose ODEAction, a chunk-free action generation paradigm that replaces action-chunk prediction with Continuous Action Evolution. Given visual observations, language instructions, and robot proprioceptive information, ODEAction learns an action transition field defined over action space, where a transition vector characterizes the evolution rate of the current action with respect to evolution time. During execution, the model takes the current action as its initial condition and integrates the learned field using an ordinary differential equation (ODE), recursively generating subsequent actions with explicit dependencies. ODEAction is implemented as an action head and does not depend on a specific vision-language backbone architecture. Experiments on LIBERO, EBench, RoboTwin, and real robots show that ODEAction improves both task performance and trajectory quality over matched action-chunking baselines, with particularly pronounced gains on fine manipulation tasks.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.