An Approach to Estimate Robot Pose under Uncertainty

Estimating robot pose accurately is a key problem in incremental mapping for unknown environment, because there are uncertain information both in the map that has been partially built and in the data from sensors. This paper proposes an algorithm for pose estimation under those uncertainties based on Cox algorithm, which reckons robot pose with the requirement that the map is totally known and exact. The way matching an image to a model is used to get a rude correspondence between the sensor data and the half-baked map. Then removing improper matching and employing weighted matrix are applied to reduce the impact of data´s and map´s error. The refined pose is estimated by weighted least square with the best congruence. In addition, the error resulted from pseudocongruence are resolved by adding virtual lines and dots. Experimental results from actual runs demonstrate that our approach is well suited for pose estimation under uncertainty