LeWAM’s researchers report 89.7% success versus LeWM’s 28.6% on simulated contact-rich tasks, with 32.3× faster planning; code and checkpoints are public.