A Reliable and Low Cost Vehicle Localization Approach Using Interval Analysis

Abstract

Vehicle localization is a part of many automotive applications where reliability is of crucial importance. Probabilistic techniques such as Kalman and Particle filtering have long been used to solve robotic localization and mapping problem. Despite their good performance in practical applications, they could suffer inconsistency problems (unreliable results). This paper proposes an interval analysis based method to estimate the vehicle pose (position and orientation) in a consistent way, by fusing low cost sensors and map data. With bounded-error parametric models, we cast the problem into an Interval Constraint Satisfaction Problem (ICSP), solved via Interval Constraint Propagation (ICP) techniques. An interval map is built when a vehicle embedding expensive sensor navigates around the environment. Then vehicles with low cost sensors (dead reckoning and monocular camera) can use this map for reliable ego-localization. Experimental results show the soundness of the proposed method in achieving consistent localization.