AEGIS: Anchor-Enforced Gradient Isolation for Knowledge-Preserving Vision-Language-Action Fine-Tuning

Excerpt

arXiv:2604.16067v2 Announce Type: replace Abstract: Fine-tuning pre-trained Vision-Language Models (VLMs) for robotic manipulation introduces a fundamental stability-plasticity dilemma: continuous flow-matching action experts backpropagate concentrated, low-rank regression gradients into transformer backbones trained on high-dimensional cross-entropy objectives. This cross-modal gradient asymmetry rapidly degrades pre-trained visual reasoning. Existing solutions either disconnect continuous grad