Goals

• [x] Calibrate the sensors in gazebo
• [x] Implement detection/classification through cameras

Calibration

GitHub - lardemua/atom: Calibration tools for multi-sensor, multi-modal robotic systems

ATOM was used to calibrate the sensors, since it has a lot of tools for multi-sensor systems which is the case of the Atlascar2.

Problems faced

• Charuco pattern in the Atlascar2 repository has wrong configuration of the axis for the ATOM calibration process. The Z axis of the pattern points towards the camera when facing each other. This causes the pattern to be non visible from the sensor’s point of view during calibration which leads to big reprojection errors.
• The configuration of the calibration package gets the transformations of the frames from the robot description xacro file by default. Since world frame of Atlascar2’s system is Odometry which is not defined in that file, there’s a need to get the transformations from the rosbag file instead.
• A rosbag with different poses of the pattern is needed to get collections. For this purpose, ATOM provides a tool in Rviz called interactive marker which allows the user to move and rotate the pattern in Rviz. There were some issues that were addressed in:

Interactive pattern wrong reference frame · Issue #45 · lardemua/atlascar2

To tackle this problem two scripts were created:

• pattern_pose_subs.py - allows the user to store in a list poses of the pattern by clicking ‘p’ keyword.
• pattern_pose_pub.py - publishes those poses with a time interval defined by the user.

The idea is to start recording the rosbag and right after run the pattern_pose_pub. This way there’s no time wasted which results in rosbags with smaller sizes.

Results

The results were good as expected since it’s a simulated environment. The cameras and the lidar had sub pixel and millimetre accuracy, respectively:

To check if the calibration was working well some noise was added to the initial positions of the sensors which increased the errors. Eventually, they ended up converging to similar good results.