Robot calibration of sensor poses and region based odometry using offline optimisation of map information

The calibration of mobile robot odometry and sensor extrinsic parameters can significantly improve the accuracy of robot mapping and localisation. This paper reports on the simultaneous calibration of wheeled robot odometry and its onboard sensor extrinsic parameters. The calibration is achieved with the offline optimisation of map quality using a particle swarm optimisation algorithm. The approach takes advantage of a state-of-the-art map quality metric for 2D occupancy grid maps and uses this quality measurement as the fitness value for the particle swarm optimization algorithm. No ground truth map is required. Since odometry calibration depends on the floor surface type, the paper improves previous calibration efforts by employing a novel floor colour classification system. Floor classification is based on a Support Vector Machine that achieves greater than 98% precision and recall values for a testing dataset consisting of six different floors. We demonstrate the benefit of our proposed floor classification system in the calibration of odometry and sensor extrinsic parameters by real world experiments. Furthermore, the consistency of our calibration method is also validated experimentally with different data sets.