Robust self-calibration

Motivation and scope

Calibration of robot sensors is essential to create safe and efficient robot systems. Extrinsic sensor calibration is the process of finding the position and orientation offset between robot and sensor coordinate frames. The calibration process typically involves tracking the motion of the sensor e.g. based on point cloud registration. Previous methods for extrinsic calibration are unable to introspectively assess the quality of the calibration. Consequently, the calibration procedures are unable to understand when calibration has succeeded or failed. In this work we intend to address the problem of robust calibration by utilizing state-of-the-art quality assessment.

Specific tasks 

In this project, the task is to calibrate the offset between robot and velodyne 3d lidar sensor. Ros nodes for calibration and quality assessment are provided and needs to be integrated into a robust calibration system. Experiments will be done on an existing truck with ros infrastructure within a warehouse environment.

Calibration procedure. The program finds a correct alignment at 0:16, after which the blurry map starts getting crisper.

Presentation of a novel scan alignment measure that may be used to detect calibration errors.

Necessary skills

C++/Python and ROS.

Benefits

After this work you will have gained insight and tools to address core perception/localization challenges faced in virtually all robotic projects. 

Contact

Daniel Adolfsson