Exploration vs. a human pointing
random exploration
human demos
reset
random rollout
human demo / policy fit to demos
policy tube (±2σ of demos)
random exploration
tries
0
reached the cup
0
hit rate
–
5 demos + policy
tries
0
reached the cup
0
hit rate
–
How big is the space?
joints
50
task length
10 s
control rate
50 Hz
values / joint / step
10
possible trajectories