Research On Local Path Planning Of Mine Rescue Robot Based On Neural Network

The mine environment structure is complex and disaster accidents occur frequently.In the process of underground rescue,the complex environmental structure seriously affects the search and rescue efficiency of rescuers,and the occurrence of secondary disaster accidents seriously endangers the life safety of rescuers.Therefore,the development of mine rescue robots is important for improving the efficiency of underground search and rescue and avoiding injuries to rescuers.Path planning is a key technology for rescue robots to launch search and rescue independently,and the response speed of the local path planning algorithm has a great impact on the rescue efficiency of rescue robots.In this thesis,we propose a neural network-based local path planning method for mine rescue robots to improve the rescue speed and search efficiency of rescue robots and address the problem of slow response time of existing local path planning algorithms for mine rescue robots.Using the path data of global path planning as a guide,a neural network model that can be used for local path planning of rescue robots is established,and on this basis,it is fused with the global path planning algorithm.The main research contents of this thesis are as follows:(1)For the problem of too many redundant points in the paths planned by the traditional Dijkstar algorithm,an improved Dijkstar algorithm based on the key node selection strategy is proposed.The raster map modeling method is selected to establish the map model of the rescue robot working environment,based on which the traditional Dijkstar algorithm is used for path planning to obtain the basic search path.Then,for the problem of too many redundant points in the path,the key node selection strategy is used to eliminate the redundant common nodes and turning points in turn,and only key nodes such as the starting point,some necessary turning points and the end point are retained.The simulation results show that the path length planned by the improved Dijkstar algorithm is reduced by 4.9%,the number of nodes is reduced by90.6%,and the number of turning points is reduced by 59.1%.(2)The trajectory optimization method based on the improved Bezier curve is studied for the problem of unsmooth global path curvature.In order to solve the problem that the path after Bezier curve optimization cannot effectively fit the original path and easily collide with obstacles,a Bezier control point selection strategy based on the global path turning point is proposed.On this basis,the second-order Bezier curve is used to optimize each set of control points separately,and the global path optimization is achieved by combining multiple Bezier curves.The simulation results show that the global path optimized by the improved Bezier curve method fits the original path more closely and effectively avoids the problem of collision with obstacles.(3)For the problem of slow response speed of traditional local path planning algorithm,a neural network-based local path planning model construction method is studied.Combining the characteristics of the obstacle environment in the global path,a feature data set collection method is proposed,and a training data set is constructed using the improved global path planning method of this thesis.Based on this,a local path planning model based on a six-layer neural network is trained.To address the problem of low success rate of path planning of this model,the influence of different number of obstacle features on the planning effect of the model is analyzed,and the feature dataset acquisition method is optimized to improve the success rate of path planning of the model.The simulation results show that the local path planning model of neural network established in this thesis has faster computation speed and shorter path length compared with the DWA algorithm.(4)For the problem that the local path planning model cannot satisfy the rescue robot to conduct global search,the fusion method of global path planning and local path planning algorithm is studied.Combining the characteristics of different path planning algorithms,the algorithm fusion method is determined.The A* algorithm is used as a guide,and the algorithm fusion is carried out with the local path planning model of neural network proposed in this thesis,and different guide point selection strategies are formulated for different obstacle situations around the rescue robot to achieve the fusion of the two algorithms.The simulation experimental results show that the fusion algorithm is able to carry out effective obstacle avoidance with the global path as the guide in the mine environment,and complete the path planning in the environment of unknown obstacles from the starting point to the end point on the basis of ensuring the global optimum.