Unleashing More Actions via Action Compositional Training for VLA Models

2026-06-30

AI Summary

Existing vision-language-action (VLA) models are prone to overfitting specific behavioral patterns during training, limiting their ability to generalize to out-of-distribution tasks requiring novel action compositions. This work proposes ACT-VLA, a framework that, for the first time, enables offline data augmentation without additional human-collected demonstrations. By uncovering latent task representations within the model, ACT-VLA automatically synthesizes physically plausible new demonstration trajectories, thereby expanding the training distribution. The approach substantially enhances the compositional generalization of VLA policies to unseen tasks, significantly outperforming baseline methods in simulated environments and demonstrating both its effectiveness and scalability.

Abstract

Vision-Language-Action models excel at robotic manipulation, driven by the scale and diversity of demonstration data. However, standard training paradigms often cause VLA models to severely overfit to specific behavioral patterns, rendering them unable to generalize to out-of-distribution scenarios even when those scenarios merely require novel combinations of identical sub-skills. While expanding datasets can mitigate this overfitting, acquiring high-quality robot data remains notoriously labor-intensive and cost-prohibitive. To resolve this impasse without expensive human teleoperation and to truly unleash more actions,i.e., enable VLA models to compose known sub-skills into a much broader set of executable behaviors beyond the original demonstrations-we propose ACT-VLA (Action Compositional Training for VLA Models), an offline data augmentation framework that leverages the model's latent task representations to synthesize novel, physically valid demonstrations directly from existing tasks for policy training. By eliminating additional manual data collection, our method automatically expands the training distribution and mitigates overfitting. We evaluate our approach on challenging manipulation tasks in simulation. Experiments demonstrate that while baseline VLA models generalize poorly due to original distribution overfitting, policies trained with our synthesized data achieve substantially higher success rates, validating that leveraging existing tasks for automated demonstration synthesis provides an effective, scalable, and data-efficient route to broadening VLA generalization.

Problem

• Vision-Language-Action models
• overfitting
• out-of-distribution generalization
• action composition
• data efficiency

Innovation

• Action Compositional Training
• VLA Models
• Offline Data Augmentation
• Demonstration Synthesis
• Generalization