WARNING - this code requires several other ROS nodes to run - currently UNDER CONSTRUCTION
This repository contains tools for performing eye-hand calibration on the KUKA arm with PTU/Kinect head. It consists of two tools:
This tool is a control script, eyehand.py, that commands the KUKA arm in a grid with an Aruco tag calibration plate as its end effector. In all target poses, the calibration plate remains parallel to the camera.
This control script delegates the data collection task to a C++ observation node. This node collects as much state as possible about the position of the arm at every position, to be used for later calibration processing.
This tool performs several post-processing tasks on the raw data collected above. Namely, it computes the average location of the arm at each point in time, as measured by both the AR tags and forward kinematics, and cleverly crops the point cloud down to include only the points from the calibration plate. It also computes a combined point cloud from all frames of a given datapoint, and performs an ICP fit of a plane to the data.
Running "go.sh calibration" in momap_descriptions will generate the correct URDF for calibration
The eye_hand_calibration repo (ROS package) is in archive.tgz.
To run the data collection again, use these commands after initalizing the robot drivers and the kinect perception stack: roslaunch eyehand_calib calib_stack.launch rosrun eyehand_calib eyehand.py
Saved IK solutions: full_calib_ik.save Saved trajectory solutions: full_calib_traj.save (Count: 61) Saved URDF: 20170817T183837_kinect2_calib_ir.urdf
Each folder and raw data CSV file corresponds to a unique head/arm pose. This corresponds to the "datapoints" referenced later in this document.
At each datapoint, several frames were collected. Details of frame colletions are shown below. An approximate time filter was used to gather all data simulataneously to within a small margin of error.
- 20170816T160945_kinect2_calib-merged.csv contains all the collected data except point clouds and images, organized by datapoint and frame.
It was generated using the command: find . -name "*_raw_data.csv" | cut -b 3-5 | xargs -I{} sed 's/^/{},/' {}_raw_data.csv > 20170817T183837_kinect2_calib_ir-merged.csv
The columns are: 0 : Datapoint 1 : Frame 2 : Timestamp 3 : Desired Pan of head joint 4 : Desired Tilt of head joint 5-11 : Desired IIWA joint angles, ordered 0-6 12 : Actual Pan of head joint 13 : Actual Tilt of head joint 14-20 : Actual IIWA joint angles, ordered 0-6 21-27 : AR tag location, in camera frame 28-35 : AR tag corner pixel locations (CW from top right) 36-42 : AR tag top left corner location, in camera frame 43-49 : AR tag top right corner location, in camera frame 50-56 : AR tag bottom left corner location, in camera frame 57-63 : AR tag bottom right corner location, in camera frame