Path planning for bearing-only simultaneous localisation and mapping

Simultaneous localisation and mapping (SLAM) is the process of estimating the pose of a mobile robot and the locations of landmarks by using sensors. When SLAM is cast as an information extraction procedure, its quality can be defined as the amount of uncertainty contained in the resultant estimation. Due to the characteristic of the bearing-only sensor and the geometry of the environment, the estimation uncertainty relies critically on the amount of information obtained from measurements and the efficiency of information extraction by the estimator. These quantities are dependent on the relative position between the robot and the landmarks, i.e., the path of the robot motion. Therefore, a well planned path of motion for the robot can significantly improve the SLAM quality. A genetic algorithm is adopted in this research to design a near-optimal one-step-ahead robot path subject to a multiple of planning objectives. The use of genetic algorithm together with a Pareto set, is proved to be efficient in reducing the estimation uncertainty and improving the quality of SLAM by simulation results.