Feature estimation and parameter tuning using H∞ Filter on FastSLAM framework

Simultaneous Localization and Mapping (SLAM) plays a key role in enabling mobile robots to autonomously navigate and build a map of their environment. Despite its wide range of applications, localization and mapping has its own disadvantages which include computation, feature representation,sensor noise, data association and robustness in dynamic environments. EKF-SLAM and FastSLAM are two frameworks of SLAM. However, they face problems related incorrect priori knowledge of mean, correlation and covariances of process as well as measurement noise as Kalman Filter acts as a minimum covariance estimator. In this paper an Extended H-infinity based FastSLAM is proposed which takes into consideration worst case minimization error and priori knowledge of noise is not essential. The proposed framework is an integration of Rao-Blackwellization Particle filter for proposal distribution and robot pose estimation along with Extended H-Infinity Filter for feature/landmark estimation. Further, the impact of the tuning parameter θ on accuracy is tested. The proposed algorithm is evaluated on a simulated environment and is compared with the existing frameworks.