Announcement_17
Recent result: standard offline actor-critic post-training took a billion-scale flow-matching VLA from a 4% behavior-cloning floor to 94% task success in simulation—without a flow-specific actor objective.
Recent result: standard offline actor-critic post-training took a billion-scale flow-matching VLA from a 4% behavior-cloning floor to 94% task success in simulation—without a flow-specific actor objective.