Research On Mapping And Path Planning Of Coal Mine Rescue Robot

China is a super country of production and consumption of coal mine as well as coal mine accidents prone country. The coal mine rescue robot has great significance to the mine disaster rescue. Because the autonomous performance of the mine rescue robot is not only one of the important indicators to measure the intelligence level of the robot, but also one of the essential performances to assist rescue teams to complete the rescue work efficiently and reliably. This paper takes the coal mine rescue robot autonomous performance as research background, uses the CUMT2, which is developed by the Research Institute of rescue equipment and technology, China University of Mining and Technology as a platform, focuses on two difficulties and hot spots around the autonomous robot: SLAM and path planning. The main researches are as follows:The hardware and software platform of the CUMT2 robot is designed according to the autonomous performance of robot. We transplanted Robot Operating System(ROS) into the robot and designed the software architecture of the robot, which consists of five main components: chassis control system, coordinate transformation system, graphical user interface, SLAM system and path planning system. We chose the hardware part of the hardware system and designed the hardware implementation of CUMT2’s drive system and perception system. The Markov assumptions and statistical methods are introduced due to the sensor noise of robot and the uncertainty of the external environment to establish the probability model of the robot. In particular, we established two kinds of robot motion models for drive system, which are velocity model and odometer model. We also established three kinds of perception model based on the perceptual system: laser beam model, likelihood model and feature-based perception model.Research on localization algorithm based on a known map due to the localization problem of the robot. First of all, the principle of the Bayesian filter is derived, and we investigated Kalman filter(KF), extended Kalman filter(EKF) and particle filter(PF) respectively, which are both variants of Bayesian filters. Then we derived the principle of the of the localization algorithm based on EKF and PF respectively. Finally, we carried out the experiment of the localization algorithm based on scan matching. The experimental results show that the introduction of sensor information as reference can improve the localization accuracy and the rate of convergence of the algorithm in the initial position and orientation estimation.Research on mapping algorithms based on known localization and unknown localization due to the mapping problem of the robot. In the mapping algorithm, we mainly investigated the mapping algorithm based on the forward perception model and the inverse perception model. The principles of the two algorithms are derived and their respective advantages and disadvantages are discussed. Compared to the mapping algorithm based on the forward perception model, the mapping algorithm based on the inverse perception model has low map consistency and high computational efficiency. The above two kinds of algorithms can obtain the ideal effect to CUMT2 robot which equipped with a laser sensor.On the SLAM algorithm research, the SLAM algorithm based on EKF, the SLAM algorithm based on Rao-Blackwellized particle filter(FastSLAM) and the SLAM algorithm based on scan matching(Hector_SLAM) are investigated respectively based on the research on the localization and mapping algorithm above. First of all, the principle of the SLAM algorithm based on EKF is deduced and simulated. Simulation results show that the accuracy of SLAM is proportional to the number of landmarks when the increase of landmarks does not introduce ambiguity between them, also the estimation accuracy of the pose for robot and landmarks will be improved when there is loop path. Secondly, the principle of FastSLAM 1.0 algorithm based on the fusion of the odometer is studied. And we also research on FastSLAM 2.0 algorithm, which solved the problem of particle degeneracy due to the mismatch between perception model and motion model by the optimization of sampling density function. This algorithm is used in building the indoor and laboratory corridor map. Experimental results show that the FastSLAM 2.0 algorithm can achieve high accuracy and consistency of the map in the case of effective odometer information. Finally, the principle of Hector_SLAM algorithm is derived, which does not need the odometer information due to the bumpy road in the underground of coal mines. And we also improve the algorithm by fusing the AHRS sensor information, which is used in building the indoor and laboratory corridor map and compared with the Fast SLAM 2.0 algorithm. Experimental results show that the improved Hector_SLAM Algorithm is applicable for the bumpy road in the underground, which can achieve high map accuracy and consistency without odometer information. But compared with the FastSLAM 2.0 algorithm, it still has the space to improve.Research on robot path planning algorithm according to the autonomous driving problem of the robot. First of all, the global path planning and local path planning for autonomous robot are studied respectively. The principle of the A* algorithm for global path planning is deduced and simulated. Simulation results show that the algorithm can plan the optimal collision free global path while taking into account the computational efficiency. The principle of the DWA algorithm for local path planning is deduced and simulated. Simulation results show that compared with the traditional potential field method, the algorithm introduces the dynamic constraints of the robot in local path planning. The running speed of the robot can be reduced dynamically due to the number of the obstacle which can effectively improve the safety of the robot. Secondly, we research on the path planning system based on layered costmaps, layered costmaps model of CUMT2 robot is proposed to solve the problem of the single layer costmap model. We realize the organic combination of the two path planning problems by the layered costmaps, and we also constructed the robot path planning system. Finally, we conducted experiments to analyze the path planning and SLAM algorithm in the laboratory corridor and underground simulation tunnel. Experimental results show that the CUMT2 robot can autonomous driving and avoid dynamic obstacles based on path planning algorithms. The map, which is constructed by the autonomous robot, has high accuracy and consistency and achieves the desired results...