IMLE-VLA is claimed to generate action chunks at 55 Hz
A researcher on the IMLE-VLA team says the VLA policy replaces flow matching with conditional IMLE and produces over 2.2 times less jerk.
TLDR
A researcher introducing IMLE-VLA says it generates action chunks at 55 Hz, 3.67 times faster than ฯ0.5. They also report over 2.2 times less jerk and task completion more than 3.9 times faster.
Combined views
116
1 Source, first seen 11h ago
31 reposts
