A ROS package for bin-picking using AFFGA-Net with ur5e.

(1) Robot: UR5e

(2) Camera: Realsense D435i

(3) Gripper: Robotiq 140-2f

(4) Notice: When the robot arm is in the watch position, the camera is ==0.6m== away from the table.

(1) Download pretrained weight affga_d.zip from https://drive.google.com/file/d/1wHYOYwWoLBAylW3Yf0NWJ60QbRMT_1zz/view?usp=drive_link, unzip it and place at scripts\grasp_methods\affga_net\pretrained_model\affga_d.

(2) Modify camera parameters camera_k at scripts\detect_grasps.py.

(3) Modify hand_eye_calibration launch at launch\grasp_affga.launch, the file must publish the coordinate system of the camera camera_color_optical_frame with respect to the robot base, or you can change the coordinate system name to your own at scripts\policy.py (Line 73).

(4) install robotiq package: refer to https://github.com/dexin-wang/robotiq_2f_gripper_control.

(1) run robot, camera, robotiq, hand_eye_calibration:

roslaunch AFFGA_binpicking_ur5e grasp_affga.launch

(2) run policy and grasp detection:

rosrun AFFGA_binpicking_ur5e policy.py
rosrun AFFGA_binpicking_ur5e detect_grasps.py

==Notice:==

In our setting, hand_eye_calibration uses py2, policy.py and detect_grasps.py use py3, so run them separately.