0 / 60 seg.

Now we take those human demonstrations -- they're all noisy and imperfect -- and we extract from them an inferred task trajectory and control sequence for the robot.