Photogrammetric calibration of mobile robot kinematics

Abstract

Numerous simulation and control applications of mobile robotics require accurate kinematic models. A kinematic model relates the position and orientation of a robot to control inputs. This thesis proposes a non-intrusive methodology to calibrate kinematic models for wheeled mobile robots. Model calibration requires accurate measurement of kinematic state and robust estimation techniques to account for process and measurement uncertainty. A photogrammetric camera system is used to measure the kinematic trajectory of wheeled mobile robot. A fully projective formulation of Lowe's pose recovery algorithm is used to estimate robot pose from monocular images. A derivative free form of the extended Kalman filter is applied to the time series pose data to estimate robot model parameters. Experimental results are presented for a differential wheeled mobile robot. Calibration by photogrammetry is shown to be viable for typical mobile robot applications.