Towards Human-like Physical Intelligence: LifelongVision-Language-Action Learning for Robotic Manipulation

2026-07-16

AI Summary

This work addresses the plasticity–stability trade-off faced by robots when continuously learning new manipulation tasks in open-world environments. To this end, we propose LifelongVLA, a framework that integrates a dual-timescale LoRA gating module with a task-aware gate to explicitly modulate the model’s adaptability. Additionally, we introduce a cache-efficient stochastic skill replay strategy to consolidate previously acquired knowledge. Built upon LoRA-based fine-tuning and a dual-path adapter architecture, LifelongVLA significantly enhances both the acquisition of novel skills and the retention of previously learned tasks on the xArm platform, thereby substantially reducing the need for costly retraining during real-world deployment.

Abstract

Similar to the natural capabilities of humans to sequentially learn new tasks, robots with Vision-Language-Action (VLA) models should possess lifelong learning ability to learn a new task when deployed in open-world environments. However, most recently proposed lifelong learning models aim to effectively learn the current task (plasticity) or maintain high accuracy on previous tasks (stability), while the plasticity-stability trade-off remains largely unsolved in robotic manipulation models. To address this fundamental challenge, we propose a cache-efficient lifelong Vision-Language-Action learning framework for robotic manipulation (i.e., LifelongVLA), which alleviates the plasticity-stability trade-off with a dual-timescale adaptation mechanism while achieving low-cost robotic deployment with a cache-efficient replay strategy. More concretely, we propose a dual-timescale LoRA gating module to decompose VLA adaptation into two lightweight pathways: a short-term adapter for plasticity and a long-term adapter for stable consolidation. These pathways are integrated via a task-aware gate, enabling explicit control of the plasticity-stability trade-off. In the skill replay phase, a cache-efficient stochastic replay strategy is proposed to preserve more balanced retention signals without full-trajectory storage. Finally, experiments show that LifelongVLA outperforms existing baselines, demonstrating efficient skill expansion, robust retention of learned manipulation behaviors, and reduced reliance on retraining for real-world deployment on an xArm robot.

Problem

• lifelong learning
• plasticity-stability trade-off
• robotic manipulation
• Vision-Language-Action
• open-world environments

Innovation

• lifelong learning
• Vision-Language-Action (VLA)
• plasticity-stability trade-off
• dual-timescale adaptation
• cache-efficient replay