#非马尔可夫 (1)

FM-VLA: Force-based Memory for Vision-Language-Action Models in Contact-Rich Manipulation