A constrained SLAM approach to robust and accurate localisation of autonomous ground vehicles

摘要

For driving assistance systems, intelligent vehicles and autonomous robots to be viable in complex environments, it is necessary to have a reliable and robust localisation function. Due to the large variability and uncertainty of such complex environments, which include theme parks, university campuses, suburbs, industrial estates and the like, it is difficult to rely on a specific method or set of sensor data to correctly and robustly estimate the robot path/pose. The key to solving the localisation problem is to optimally use and fuse all useful sources of information available to the mobile platform. For the envisaged environment, it is not unusual to have approximate digital maps of the road network. In this paper, in addition to the typical sensory information provided by extereoceptive and proprioceptive sensors, it is shown how a priori approximate knowledge available in the form of a road map can be systematically fused within a Simultaneous Localisation and Map Building (SLAM) framework to obtain more accurate and robust localisation results. This reformulation of SLAM through the introduction of constraints in the form of a priori map information not only makes the problem theoretically more correct in the sense of observability but also makes the system viable and effective, yielding more accurate results. The results obtained in an actual environment are presented to validate the claims.