Research On Path Planning Of Mobile Robots Based On Area Segmentation

In recent years,mobile robots have gradually penetrated into all aspects of people’s lives,thus higher requirements have been placed on the path planning algorithm in complex scenarios:robots’ working fields have been extended to large places such as airports and shopping malls.However,existing search-based path planning algorithms have low efficiency and poor real-time performance,while sampling-based path planning algorithm suffers poor initial path quality and large amount of time-consuming to converge to the optimal path.Therefore,a path planning algorithm which could quickly obtain high-quality path in large scenarios is necessary.What’s more,traditional path planning algorithms have to execute on a pre-established environment map,but in some special scenarios(such as disaster relief),it is unrealistic to build the environment map in advance,thus the robot need to be able to plan a rough path based on raw images(e.g.,aerial map,satellite map).Learning-based path planning algorithms can plan on raw images,but they requires high hardware and their efficiency is affected by map size.Therefore,path planning algorithms that can run efficiently on large-size images are required.In view of the above two problems,the main work of this thesis is as follows:To deal with the efficiency of path planning algorithms in large scenarios,a path planning algorithm based on Voronoi Area Segmentation(VS)is proposed.The basic idea of VS is to limit the exploration of irrelevant areas in the robot’s work area,and divide the areas that need to be explored into multiple smaller sub-regions to reduce the scale of the problem.VS plan paths in parallel in sub-regions to improve the search efficiency,and then optimize the path to improve path quality and save convergence time.Simulation experiments in five maps and robot experiments in office building show that compared with other path planning algorithms,VS algorithm can reduce the exploration of irrelevant areas,save the path convergence time and quickly obtain high-quality paths in large-scale scenarios.Aiming at the low efficiency and high hardware requirements of learning-based path planning algorithms on large-size raw images,Topological Neural A*(TNA*)was proposed.Based on the idea of focusing on important areas,TNA*focuses on the areas that the path may pass through.By pre-processing the original image and constructing a topological map from it,TNA*can transform the path planning problem on the largesize image into multiple path planning problems in the corresponding areas of the node on the topological map,which can lower the hardware requirements and improve the efficiency of the Neural A*due to the small size of the maps corresponding to topological nodes.Simulation experiments in the grid maps and in the raw images prove that TNA*is more efficient than other algorithms and can cope with path planning problem in special scenarios.The two algorithms proposed in this thesis can efficiently obtain feasible paths on grid maps and raw images,which verifies that the idea of area segmentation could reduce the scale of path planning problems and improve the efficiency of path planning,moreover,could solve the problem of low efficiency of robot path planning on largescale grid maps and large-size raw images.