Skip to content

Latest commit

 

History

History
172 lines (118 loc) · 4.33 KB

README.md

File metadata and controls

172 lines (118 loc) · 4.33 KB
Raw

grasp_prediction_arc2017

  • Object label and grasp affordance segmentation learned with instance image stacking
  • Picking folded objects (e.g., books) with the Suction Pinching Hand

Installation

With ROS

Install jsk_apc.

(At Your Own Risk) Without ROS (Use Anaconda)

make install  # Python3
# make install2  # Python2
Usage
# Don't do the following with soursing ROS setup.*sh
source .anaconda/bin/activate
python -c 'import grasp_prediction_arc2017_lib'
Testing
make lint

Examples

Without ROS

First, you must activate python environment:

source ~/ros/ws_jsk_apc/devel/.private/grasp_prediction_arc2017/share/grasp_prediction_arc2017/venv/bin/activate
# Or if you want to use Anaconda environment:
# source .anaconda/bin/activate  # Don't do this with soursing ROS setup.*sh
Training CNNs
Requirements
  • cupy 
    pip install cupy-cuda92==6.7.0  # Please specify a package corresponding to your CUDA version (cupy-cuda90, cupy-cuda92, ...) and the same version as your chainer version (6.4.0, 6.7.0, ...)
Object label and grasp affordance segmentation with ARC2017 dataset
cd examples/grasp_prediction_arc2017
./train_fcn32s.py -g 0 -d -p wada_icra2018
Object label and grasp affordance segmentation with book dataset
cd examples/grasp_prediction_arc2017
./train_fcn8s.py -g 0 -d -p hasegawa_iros2018  # or hasegawa_mthesis

With ROS

Setup

rosrun grasp_prediction_arc2017 install_data.py

Execution

roslaunch grasp_prediction_arc2017 sample_fcn_object_segmentation.launch

ARC2017 demonstration

In execution flow of pick task imitating ARC2017 competition, execute

roslaunch grasp_prediction_arc2017 setup_for_pick.launch

instead of

roslaunch jsk_arc2017_baxter setup_for_pick.launch

Video (Click Below)

Hasegawa IROS2018 Demo: Pick and Insert Books

Setup

rosrun grasp_prediction_arc2017 install_hasegawa_iros2018

Execution

roslaunch grasp_prediction_arc2017 baxterlgv7.launch
roslaunch grasp_prediction_arc2017 setup_for_book_picking.launch hasegawa_iros2018:=true
roslaunch grasp_prediction_arc2017 book_picking.launch json_dir:=`rospack find grasp_prediction_arc2017`/json_dirs/hasegawa_iros2018/ForItemDataBooks6/layout1

Video (Click Below)

Citation

Webpage of the paper

@INPROCEEDINGS{hasegawa2018detecting,
  author={S. {Hasegawa} and K. {Wada} and K. {Okada} and M. {Inaba}},
  booktitle={2018 IEEE/RSJ International Conference on Intelligent Robots and Systems},
  title={Detecting and Picking of Folded Objects with a Multiple Sensor Integrated Robot Hand},
  year={2018},
  pages={1138-1145},
  doi={10.1109/IROS.2018.8593398},
  ISSN={2153-0866},
  month={Oct.}
}

Hasegawa Master Thesis Demo: Grasp Books with Low Lifting

Setup

rosrun grasp_prediction_arc2017 install_hasegawa_mthesis

Execution

# Experiments of Grasp Stability
roslaunch grasp_prediction_arc2017 baxterlgv7.launch
roslaunch grasp_prediction_arc2017 setup_for_book_picking.launch hasegawa_mthesis:=true
roslaunch grasp_prediction_arc2017 book_picking.launch main:=false json_dir:=`rospack find grasp_prediction_arc2017`/json_dirs/hasegawa_mthesis/ForItemDataBooks8/each_obj/alpha_cubic_sport_wallet
roseus `rospack find grasp_prediction_arc2017`/euslisp/hasegawa_mthesis/pick-book-eval.l

# In Euslisp Interpreter
(pick-book-eval-init :ctype :larm-head-controller :moveit t)
(pick-book-eval-mainloop :larm)
## Please see warn messages and source codes for optional settings