Comparative Experiments of Moving Horizon Estimation based SLAM in Indoor Environment

In this paper, we describe accuracy verification of simultaneous localization and mapping (SLAM) in a crowded area where measured data often becomes insufficient. To realize autonomous locomotion of robots in an unknown environment, they need to create an environment map while estimating self-pose with measured data. When the robot works in an area where many people exist, however, the measured data is frequently occluded by them and become insufficient to achieve the SLAM. Applying moving horizon estimation to SLAM, when the measured data is insufficient, the evaluation of the motion model becomes more dominant than that of the matching error of the measured data. To verify the effectiveness of SLAM based on MHE, we conduct experiments in the crowded area to compare the result with that of SLAM based on extended Kalman filter. Also, we applied sequential quadratic programming to the optimization to reduce the computational complexity of the proposed method. It was confirmed that the computational complexity was reduced by about 88 % compared to the trust region method.