ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ye, Wencheng, Wang, Tianshi, Zhu, Lei, Li, Fengling, Yang, Guoli, Shen, Hengtao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917402553352192
author Ye, Wencheng
Wang, Tianshi
Zhu, Lei
Li, Fengling
Yang, Guoli
Shen, Hengtao
author_facet Ye, Wencheng
Wang, Tianshi
Zhu, Lei
Li, Fengling
Yang, Guoli
Shen, Hengtao
contents Recent Vision-Language-Action (VLA) models have shown impressive flexibility and generalization, yet their deployment in robotic manipulation remains limited by heavy computational overhead and inference latency. In this work, we present ActDistill, a general action-guided self-derived distillation framework that transfers the action prediction capability of any existing VLA model to a lightweight counterpart. Unlike previous efficiency strategies that primarily emphasize vision-language correlations, ActDistill leverages action priors to guide knowledge transfer and model compression, achieving action-oriented efficiency for VLA models. Specifically, we employ a well-trained VLA model as the teacher and introduce a graph-structured encapsulation strategy to explicitly model the hierarchical evolution of action prediction. The student model, derived from the graph-encapsulated teacher, is further equipped with a dynamic router that adaptively selects computation paths based on action prediction demands, guided by hierarchical graph-informed supervision to ensure smooth and efficient evolution. During inference, graph-related auxiliary components are removed, allowing the student to execute only dynamically routed layers and predict high-precision actions with minimal computation and latency. Experiments on embodied benchmarks demonstrate that ActDistill achieves comparable or superior performance to full-scale VLA models while reducing computation by over 50% with up to 1.67 times speedup, thereby establishing a general paradigm toward efficient embodied intelligence.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18082
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Models
Ye, Wencheng
Wang, Tianshi
Zhu, Lei
Li, Fengling
Yang, Guoli
Shen, Hengtao
Computer Vision and Pattern Recognition
Robotics
Recent Vision-Language-Action (VLA) models have shown impressive flexibility and generalization, yet their deployment in robotic manipulation remains limited by heavy computational overhead and inference latency. In this work, we present ActDistill, a general action-guided self-derived distillation framework that transfers the action prediction capability of any existing VLA model to a lightweight counterpart. Unlike previous efficiency strategies that primarily emphasize vision-language correlations, ActDistill leverages action priors to guide knowledge transfer and model compression, achieving action-oriented efficiency for VLA models. Specifically, we employ a well-trained VLA model as the teacher and introduce a graph-structured encapsulation strategy to explicitly model the hierarchical evolution of action prediction. The student model, derived from the graph-encapsulated teacher, is further equipped with a dynamic router that adaptively selects computation paths based on action prediction demands, guided by hierarchical graph-informed supervision to ensure smooth and efficient evolution. During inference, graph-related auxiliary components are removed, allowing the student to execute only dynamically routed layers and predict high-precision actions with minimal computation and latency. Experiments on embodied benchmarks demonstrate that ActDistill achieves comparable or superior performance to full-scale VLA models while reducing computation by over 50% with up to 1.67 times speedup, thereby establishing a general paradigm toward efficient embodied intelligence.
title ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Models
topic Computer Vision and Pattern Recognition
Robotics
url https://arxiv.org/abs/2511.18082