Map Fusion For Multi-robot SLAM With Heterogenous Sensors

Simultaneous localization and mapping (SLAM) is one of the key technologiesin the research and application field of robotics. Many researches have been focusedin the field of single robot SLAM. However they are restricted in thosecircumstances like the big area of environments and real-time applications.Multi-robot SLAM can well satisfy such requirements and achieve the localizationand mapping goal effectively and accurately. In addition, multiple robots equippedwith heterogeneous sensors can make full use of the different sensor information toachieve a rich-information map. So it is of great significance to research onmulti-robot SLAM with heterogeneous sensors.There are some challenges in the multi-robot SLAM such as the scalabilityissues, the robots’ relative pose, and map fusion. This dissertation focuses on themap fusion problem, and proposes an efficient approach by integrating the algorithmof adaptive monte carlo localization (AMCL) and the iterative closest point (ICP);the approach can effective solve map fusion problem for multi-robot system.Because of the global property of AMCL, it does not need to know the robots’relative poses a prior. Furthermore by using this AMCL method the sensorsinformation is gathered to optimize the result of map fusion which makes theproposal method robust.Our approach was verified by performing experiments on Turtlebots. Tworobots labeled with A and B are equipped with a laser scanner and Kinect,respectively. They executed the algorithm of FastSLAM2.0to build the occupancygrid maps at different start points but in the same environment; the odometry andmeasurement information, and the best poses with timestamps were recorded duringthe processing. With the maps built, robot A was relocated in the map built by robotB by using the recorded odometry and measurement information of robot A in theAMCL process. The relocated poses of robot A were compared to the original posesby using the timestamps and covariances to calculate the approximatedtransformation matrix between the two maps. And then, two maps were fused byutilizing the ICP method based on the calculated transformation matrix. Finally,experimental results have shown the proposed approach performs efficiently.