信息论视角下的持续视觉-语言-动作对齐约束
Information-Theoretic Constraints for Continual Vision-Language-Action Alignment
AI总结：
本文提出Info-VLA框架，通过两个互补约束保持跨模态信息结构，解决持续学习中视觉语言动作对齐退化问题，实验表明其在任务保持与适应性上优于现有方法。
摘要：
当部署在开放-ended机器人环境中时，视觉-语言-动作（VLA）模型需要持续获取新技能，但会遭受严重的灾难性遗忘。我们观察到这种退化与跨模态信息结构的退化有关，其中视觉观察、语言指令和动作之间的依赖关系在持续适应过程中逐渐扩散。但现有持续学习方法未能保持此类跨模态信息依赖。因此，我们提出了Info-VLA，一种信息保持的持续学习框架，通过两个互补约束维持跨模态信息结构。Replay Anchor Contrastive Learning从冻结的教师模型中构建稳定的对齐锚点，在表示空间中保持跨模态对齐。Cross-Modal Mutual Information Maximization进一步通过互信息约束保持视觉和语言表示之间的依赖结构。通过共同保持历史对齐和跨模态依赖信息，Info-VLA在持续学习过程中平衡了稳定性和可塑性。此外，在LIBERO上的实验表明，Info-VLA在任务保持和适应性上显著优于现有方法。
英文摘要：
When deployed in open-ended robotic environments, Vision--Language--Action (VLA) models need to continually acquire new skills, yet suffer from severe catastrophic forgetting. We observe that this degradation is related to the deterioration of cross-modal information structure, where dependencies among visual observations, language instructions, and actions progressively diffuse during continual adaptation. But existing continual learning methods fail to preserve such cross-modal information dependencies. Thus, we propose Info-VLA, an information-preserving continual learning framework that maintains cross-modal information structure through two complementary constraints. Replay Anchor Contrastive Learning constructs stable alignment anchors from a frozen teacher model, preserving cross-modal alignment in the representation space. Cross-Modal Mutual Information Maximization further preserves dependency structure between visual and language representations through mutual information constraints. By jointly preserving historical alignment and cross-modal dependency information, Info-VLA balances stability and plasticity during continual learning. Furthermore, experiments on the LIBERO show that Info-VLA significantly outperforms existing methods in both task retention and adaptation.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
