Optimal global pose estimation for consistent sensor data registration

We consider the problem of consistent range data registration in modeling an unknown environment. The problem is expressed as the optimal estimation of pose variables under the maximum likelihood criterion. By treating all the history of robot poses as variables and solving them simultaneously, consistency is enforced. We formulate relative pose constraints from both matched scans and odometry measurements to construct a network of measurements. Then we derive closed-form pose estimates as well as their covariance matrices. Examples of global scan registration using both real and simulated data are presented