Shows that simple sequential fine-tuning with LoRA outperforms complex algorithms for continual reinforcement learning in VLA models.
arXiv · March 13, 2026 · 2603.11653
Why it matters
It de-bunks the assumption that large embodied models require sophisticated 'continual learning' strategies to avoid catastrophic forgetting. This simplifies the recipe for building self-improving robots, showing that scale and parameter-efficient tuning naturally stabilize the learning process.
From the abstract
Continual Reinforcement Learning (CRL) for Vision-Language-Action (VLA) models is a promising direction toward self-improving embodied agents that can adapt in openended, evolving environments. However, conventional wisdom from continual learning suggests that naive Sequential Fine-Tuning (Seq. FT) leads to catastrophic forgetting, necessitating complex CRL strategies. In this work, we take a step back and conduct a systematic study of CRL for large pretrained VLAs across three models and five c