SLAM with MTT: Theory and Initial Results

To make a robot to work for and with human, the ability to simultaneously localize itself, accurately map its surroundings, and safely detect and track moving objects around it is a key prerequisite for a truly autonomous robot. In this paper, we explore the theoretical framework of this problem, i.e. Simultaneous Localization and Mapping (SLAM) with Multiple Target Tracking (MTT), and propose to employ Sequential Monte Carlo Methods (SMCM) as robust and computationally efficient algorithm. After mathematically formulating the problem, we apply a Rao-Blackwellized particle filter to solve SLAM which is partitioned into robot pose and feature location estimations and a conditioned particle filter to solve MTT which is partitioned into robot pose and moving object state estimations, both filters conditioned on robot pose. In detail, we propose Sampling Importance Resampling (SIR) method to estimate robot pose, Extended Kalman Filter (EKF) to estimate feature location, and Hybrid Independent/Coupled Sample-based Joint Probability Data Association Filter (Hyb-SJPDAF) to solve tracking and data association problem. We present some preliminary experimental results to demonstrate the capabilities of our approach. Key words—Simultaneous Localization and Mapping (SLAM), Multiple Target Tracking (MTT), Sequential Monte Carlo Method (SMCM), Sampling Importance Resampling (SIR), Extended Kalman Filter (EKF), Joint Probability Data Association (JPDA)