Neural Manipulation Planning on Constraint Manifolds
Ahmed H. Qureshi, Jiangeng Dong, Austin Choe, and Michael Yip
University of California San Diego
The following videos show the CoMPNet path solutions in practical motion planning tasks involving both unconstrained and constrained problems. It can be seen that CoMPNet generalizes to new locations of the objects in the given environments with high success rates. Also, when compared to the state-of-the-art constrained motion planning algorithms, CoMPNet outperforms by order of magnitude improvement in computational speed with a significantly lower variance.
The following videos show the CoMPNet path solutions in practical motion planning tasks involving both unconstrained and constrained problems. It can be seen that CoMPNet generalizes to new locations of the objects in the given environments with high success rates. Also, when compared to the state-of-the-art constrained motion planning algorithms, CoMPNet outperforms by order of magnitude improvement in computational speed with a significantly lower variance.
Bartender Task
Bartender Task
In this setting the robot is required to clean the tables by placing the cans and bottle to the trash, and carefully moving (without tilting) the mug and kettle to the tray.
In this setting the robot is required to clean the tables by placing the cans and bottle to the trash, and carefully moving (without tilting) the mug and kettle to the tray.
The total mean computation time for all manipulation tasks for CoMPNet, CBiRRT, Atlas-RRT and TB-RRT are 16.52s, 167.59s, 43.21s, and 73.29s, respectively.
The total mean computation time for all manipulation tasks for CoMPNet, CBiRRT, Atlas-RRT and TB-RRT are 16.52s, 167.59s, 43.21s, and 73.29s, respectively.
Kitchen Task
Kitchen Task
In this setting the robot is required to organize the kitchen by i) placing the cans and bottle to the trash; ii) open the cabinet and carefully moving (without tilting) the mugs from inside to the tray; and iii) carefully move (without titling) the pitcher into the cabinet.
In this setting the robot is required to organize the kitchen by i) placing the cans and bottle to the trash; ii) open the cabinet and carefully moving (without tilting) the mugs from inside to the tray; and iii) carefully move (without titling) the pitcher into the cabinet.
The total mean computation time for all manipulation tasks for CoMPNet, CBiRRT, Atlas-RRT and TB-RRT are 37.78s, 236.39s, 134.78s, and 164.87s, respectively.
The total mean computation time for all manipulation tasks for CoMPNet, CBiRRT, Atlas-RRT and TB-RRT are 37.78s, 236.39s, 134.78s, and 164.87s, respectively.
CoMPNet on the Real Robot in the Bartender Task
CoMPNet on the Real Robot in the Bartender Task
Reach the red bull can and move it to the trash bin.
Reach the red bull can and move it to the trash bin.
Reach the mug and carefully move it to the tray.
Reach the mug and carefully move it to the tray.
Bibliography
Bibliography
@ARTICLE{qureshicompnet2020,
@ARTICLE{qureshicompnet2020,
author={A. H. {Qureshi} and J. {Dong} and A. {Choe} and M. C. {Yip}},
author={A. H. {Qureshi} and J. {Dong} and A. {Choe} and M. C. {Yip}},
journal={IEEE Robotics and Automation Letters},
journal={IEEE Robotics and Automation Letters},
title={Neural Manipulation Planning on Constraint Manifolds},
title={Neural Manipulation Planning on Constraint Manifolds},
year={2020},
year={2020},
volume={5},
volume={5},
number={4},
number={4},
pages={6089-6096},}
pages={6089-6096},}