Enhancing Generalization in Vision Language Action Models by Preserving Pretrained Representations

Publication
International Conference on Intelligent Robots and Systems