Improved tracking and docking of Industrial Mobile Robots through UKF vision based kinematics calibration

Performing an open-loop movement, or docking, for an industrial mobile robot (IMR), isa common necessary procedure when relying on environmental sensors is not possible. This procedureprecision and outcome, solely depend on the IMR forward kinematic and odometry correctness, which is tiedto the kinematics parameters, depending on the IMR kind. Calibrating the kinematic parameters of an IMRis a time consuming and mandatory procedure, since the mechanical tolerances and the assembly proceduremay introduce a large variation from the nominal parameters. Furthermore, calibration inaccuracies mightintroduce severe inconsistencies in tasks such as localization, mapping, and navigation in general. In thiswork, we focus on the so-called kinematic parameter calibration. We propose the use of the unscentedKalman filter to perform a calibration procedure of the geometrical kinematic parameters of a mobileplatform. The mobile platform is externally tracked during the calibration phase, using a fixed temporaryexternal sensor that retrieves the position of a visual tag fixed to the platform. The unscented Kalman filter,using the calibration phase collected data, estimates the enlarged system state, which is comprised of theparameters that have to be estimated, the platform odometry and the visual tag position.The method can either be used online, to identify parameters and monitor their value while the system isoperating, or offline, on logged data. We validate this method on two different devices, a 4 mecanum-wheelIMR , and a Turtlebot 3, using a camera to track the movement trough a reference chessboard, for thencomparing the original path to its corrected version.