#Kinematic Calibration
Kinematic Calibration is an open source calibration framework for humanoid robots. Currently, is is specifically designed for the Nao humanoid by Aldebaran.
The framework estimates the camera's extrinsic and intrinsic parameters, as well as the joint offsets of the humanoid. To this end, artifical markers (e.g. checkerboards) need to be attached to the robot's end-effectors, which are then observed by the camera.
Please note: This code is experimental and not a stable version. Parts of the functionallity might be missing or are broken. You are invited to contribute!
The approach is described in details in the paper: Daniel Maier, Stefan Wrobel, Maren Bennewitz, Whole-Body Self-Calibration via Graph-Optimization and Automatic Configuration Selection, IEEE International Conference on Robotics and Automation (ICRA), 2015
The software requires the following software:
- ROS
- hrl_kinematics (https://github.com/ahornung/hrl_kinematics.git)
- libflann
- PCL
- opencv
- g2o (only commit 96c5a7 tested)
- libgsl
- nao_robot
- opencv
- aruco (e.g. from ar_sys in ros-indigo)
- catkin
- doxygen
As this is a catkin package, simply add it to your workspace and compile with catkin.
The documention can be created with doxygen.
First of all, the framework requires that the initial marker placement is approximately known from the ROS param server. So the urdf model needs to be adjusted accordingly. The same goes for the camera calibration. There exists a nao_description package that Daniel Maier edited for nao_calibration package (https://github.com/danielmaier/nao_description). In order to simplify the process, there there exists a version of combined of both. You can simply use the combined version.
Second, a list of possible poses is required. The poses should be such that the marker is visible in the image. Such a list can be created manually (cumbersome) or using an initial calibration and the poseSampling node. It might be necessary to make some modifications in order to run that node. An auto-generated sample list is contained in the config folder (e.g. poses_larm_750.yaml) for the marker placement as illustrated in the paper. So if your markers are approximately at the same position, it should work out of the box.
Third, there are some implicit assumptions made, e.g. that you are trying to calibrate the lower of Nao's two cameras and its TF frame name is CameraBottom_frame. This assumption has been lifted in some parts of the code and the camera frame is now configurable but maybe other parts of the code still rely on this assumption. The same goes for other link and joint names.
The main config file is nao_calibration.yaml. For each kinematic chain, there is an additional config file params_$CHAINNAME_general.yaml that allows to change the name of the frames (e.g. Marker frame, Head frame, etc.), marker type (checkerboard, aruco), and some marker specific settings (dimensions). Finally, there is params_$CHAINNAME_capturing.yaml that contains capturing parameters (e.g. head step when searching for a marker)
For convenience, the following launch file should bring up all the prerequisites that are needed for the calibration:
nao_basic.launch
In details, the calibration software needs the robot description published to the parameter server (nao_description)
Furtermore, it requires (at least) the following services:
- joint_stiffness_trajectory (nao_controller)
- joint_trajectory (nao_controller)
- body_pose (nao_pose)
Also, these topics need to be published:
- camera/image_raw (nao_sensors)
- joint_states (nao_driver)
Different from Maier's code, here we have a common launch file named "common.launch", where you can run all the launch files from here and we are running the camera node from the real Nao. So before everything, run roscore on terminal and ssh your robot. After that, export your computer from robot and start the camera node on your robot.
The purpose of collecting all launch files in one file was that, there are common args in different launch files- that can be edited several times e-g- in order to change which body part to calibrate. So, rather than opening and editing several files, you can fastly edit the args from this file and decide which launch files to be launched by changing the bool values:
<arg name="use_nao_basic" default="true"/>
<arg name="use_updateNode" default="true"/>
<arg name="use_dataCaptureService" default="true"/>
<arg name="use_calibrate_nao" default="true"/>
You can again change boolean values to decide what to calibrate:
<arg name="larm_bool_c" value="false" />
<arg name="rarm_bool_c" value="false" />
<arg name="lleg_bool_c" value="true" />
<arg name="rleg_bool_c" value="false" />
If you change any of these bools, pls also adjust the config file (nao_calibration.yaml)
The updateNode writes a set of files (as sepecified in the nao_calibration.yaml file) containing the camera calibration, an updated robot model and individual joint_offset, camera transform, and marker transform files. The latter can be used in conjunction with the combined version of nao_description.
If you want to see how the calibrated version finds where the markers are, there exists a launch file:
draw.launch
It draws a red dot on where the marker is, calculated via calibrated joint data.