ReorientBot

Learning Object Reorientation for Specific-Posed Placement

Robots need the capability of placing objects in arbitrary, specific poses to rearrange the world and achieve various valuable tasks. Object reorientation plays a crucial role in this as objects may not initially be oriented such that the robot can grasp and then immediately place them in a specific goal pose.

In this paper, we present a vision-based manipulation system, ReorientBot, which consists of 1) visual scene understanding with pose estimation and volumetric reconstruction using an onboard RGB-D camera; 2) learned waypoint selection for successful and efficient motion generation for reorientation; 3) traditional motion planning to generate a collision-free trajectory from the selected waypoints. We evaluate our method using the YCB objects in both simulation and the real world, achieving 93% overall success rate, 81% improvement in success rate, and 22% improvement in execution time compared to a heuristic approach.

Overview Video (with audio)

System for specific-posed placement

ReorientBot is a hybrid of learned components (, ) and traditional motion planning, consisting of 1) vision-based 6D pose estimation and volumetric reconstruction; 2) motion waypoint generation; 3) trajectory generation with the waypoints.

Training in simulation

We train the learning models in physics simulation using the CAD models of known objects. Object pile configurations are randomly generated, for which the validity of grasp, reorientation, and regrasp are evaluated to generate training labels.

Baseline comparison for object reorientation

Our proposed method, ReorientBot, surpasses a heuristic method that uses object upright poses for reorientation, showing 93% overall success rate, 81% improvement in success rate, and 22% improvement in execution time.

Comparison of task completion. We use goal configurations that are not achievable without reorientation (146 tasks).
Method	Success % (reorient)↑	Success % (place)	Success % (overall)
Heuristic	71.9	81.0	58.2
ReorientBot	97.9	95.1	93.2

Comparison of timing. Reporting only when both methods succeeded to complete task; 38 tasks out of 146.
Method	Planning time [s]↓	Execution time [s]↓	Trajectory length [rad]↓
Heuristic	3.3	4.0	4.2
ReorientBot	2.5	3.2	3.3

Real-world experiments

Shelf storing (three cracker boxes)

Box packing (three objects)

Large workspace (many objects)

Heuristic vs. Learned (overall time)

Heuristic vs. Learned (execution time)

Paper

Bibtex

  @inproceedings{Wada:etal:ICRA2022b,
    title={{ReorientBot}: Learning Object Reorientation for Specific-Posed Placement},
    author={Kentaro Wada and Stephen James and Andrew J. Davison},
    booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
    year={2022},
  }

Contact

If you have any questions, please feel free to contact Kentaro Wada.