VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models figure
AlphaXiv 中文概览(可滚动查看)