Policies trained from data instead of hand-written controllers.
Learning fine-grained bimanual manipulation with low-cost hardware.
