This paper addresses a multi-robot SLAM approach based on the Kalman consensus filter (KCF). Under the unknown initial condition, a reference robot designates the initial poses of other robots when the first rendezvous between them occurs. Accordingly, past and current poses and maps of these robots are estimated by an acausal filter and a causal filter. | Journal of Automation and Control Engineering Vol. 3, No. 5, October 2015 Kalman Consensus Based Multi-Robot SLAM with a Rao-Blackwellized Particle Filter Seung-Hwan Lee and Beom H. Lee Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea Email: {leeyiri1, bhlee} @ they basically assumed known data association for features and the known initial condition. To perform Multi-robot SLAM, the information filter and the information consensus filter are used together in [7]. They compare the results from the information consensus filter and covariance intersection (CI). But the known conditions are still assumed in the simulation. In our previous work [8], [9], we proposed a multirobot SLAM framework. Under the unknown initial condition, robots initialize their poses when the first rendezvous with the reference robot occurs. Subsequently, the poses and maps between the N-1th and the Nth rendezvous are compensated whenever the Nth rendezvous occurs again. For the compensation, current poses for two robots are fused by Covariance Intersection (CI). Therefore, this paper presents a Rao-Blackwellized particle filter based multi-robot SLAM using the KCF in the event of rendezvous. Unlike the conventional approach, we consider several rendezvous between robots. The robots are initialized at the first meeting with a reference robot. In the case of the second rendezvous or more rendezvous, the current pose and covariance of two robots are fused via the procedure of the KCF. Based on these poses, their past poses and maps are compensated through backtracking until the most recent rendezvous point. In two simulations, we show the performance of the proposed approach in terms of the accuracy of the robot pose and map. First the conventional approach for the multi-robot SLAM framework and its problems are described in Chapter 2. In Chapter 3, the proposed approach is explained in detail. Chapter 4 shows the accuracy of the robot pose .
