Coarse-to-Fine Q-attention: Efficient Learning for
Vision-based Robotic Manipulation via Discretisation

Stephen James Kentaro Wada Tristan Laidlow Andrew J. Davison

Dyson Robotics Lab, Imperial College London
CVPR 2022, Oral

Code: https://github.com/stepjam/ARM
(Repo contains ARM and C2F-ARM)

Real World Experiments

real_saucepan.mp4

Take lid off saucepan


real_towel.mp4

Fold towel


real_car.mp4

Pull toy car


real_light.mp4

Turn on light


real_shelve.mp4

Pull cloth from shelf


Videos show the coarse-to-fine Q-attention

c2f_qattention_take_lid_off_saucepan.mp4.mp4

Take Lid Off Saucepan

c2f_qattention_put_rubbish_in_bin.mp4.mp4

Put Rubbish In Bin