Can VLA Models Learn from Real-World Data Continually without Forgetting?

2026-05-26

AI Summary

This work presents the first empirical investigation into continual learning for vision-language-action (VLA) models in real-world robotic settings, where catastrophic forgetting poses a significant challenge to retaining previously acquired skills while learning new ones. The authors introduce a realistic robot dataset comprising four sequential manipulation tasks—rigid-body grasping, contact-intensive pressing, and folding of deformable objects—and systematically evaluate the efficacy of experience replay strategies. Their experiments reveal that standard continual learning approaches suffer from substantial performance degradation on prior tasks, whereas an optimized experience replay mechanism effectively mitigates forgetting and enhances performance on both old and new tasks. These findings offer a practical pathway and critical insights for enabling effective continual learning of VLA models in real-world scenarios.

Abstract

Vision-language-action (VLA) models provide a promising foundation for general-purpose robotics. However, their successful deployment in real-world scenarios requires the ability to continually acquire new skills while retaining previously learned behaviors. While pioneering research has studied the continual learning of VLA models in narrowly simulated environments, this challenge remains largely unexplored under realistic conditions. To address this limitation, we construct a real-world continual learning dataset comprising four sequential manipulation tasks, spanning rigid-object pick-and-place, contact-rich pressing, and deformable-object folding. Using this dataset, we conduct comprehensive experiments and find that VLA models suffer significant catastrophic forgetting when continually learning from heterogeneous real-world demonstrations. We then systematically evaluate experience replay and uncover key implementation factors that govern its success. In summary, this work provides the first empirical study of real-world continual VLA learning and offers practical guidance for deploying long-lived robot policies.

Problem

• continual learning
• vision-language-action models
• catastrophic forgetting
• real-world robotics
• robot manipulation

Innovation

• continual learning
• vision-language-action models
• real-world robotics
• catastrophic forgetting
• experience replay