| With the vigorous development and gradual maturity of artificial intelligence technology,intelligent mobile agents are increasingly active in the fields of transportation and industrial production.As an important technology,path planning and tracking has always been a research hotspot and has been continuously optimized.However,there are still some problems in the planning of multi-obstacle scenes: the planning of unstructured roads is easy to fall into local optimization,which affects the search efficiency,and the path length and safety need to be further improved;In the complex scene with disordered obstacles,the path planning is difficult to balance the planning efficiency and the acquisition of the optimal path.To this end,for the above two types of multi-obstacle scenes,this paper conducts research on the planning method with high search efficiency and optimal paths for intelligent driving vehicles and mobile robots,respectively,and realizes the path tracking simulation.The main research contents are as follows:(1)An improved graph search method based on map processing is proposed for multi-obstacle scenarios of unstructured roads.Firstly,a map processing method for identifying and filling concave obstacles is proposed by drawing on the idea of rainwater harvesting,to avoid planning to fall into the local optimum.Subsequently,an improved graph search algorithm is proposed,which refines the search angle based on the 8-neighbor A~* algorithm to form the 8+4 neighborhood search.The simulation results show that the path search efficiency is increased by 5.5%,and the path length is shortened by 5.5%.At the same time,introducing the security heuristic function to solve the path and vertex collision or close to the obstacles,improve the path safety.Finally,removing the collinear points and the redundant inflection points,and smoothing the path by piecewise cubic Hermite interpolation to ensure the continuity of the path curve C1 and improve the traceability of the global path.(2)A path planning algorithm incorporating deep learning is proposed for the complex environment with disordered obstacles.Firstly,randomly select the starting and target points in the fixed environment to generate the original map,then use the4-neighbor A~* algorithm to generate the label image containing the optimal path,to complete the creation of the dataset.Secondly,the hourglass network structure is optimized to form a lightweight network fusing multi-layer feature information and determine the epoch and path pixel thresholds through model training and testing.Thirdly,the fusion algorithm is proposed,which uses deep learning to identify the region of interest and form a preprocessed map,and then uses the planning algorithm to generate the optimal path.Finally,carry out the simulation tests in Matlab.Three representative algorithms DL-A~*,DL Bi A *,and DL-RRT are selected to complete the test in two scenarios: layered path length and complete path length,with the highest efficiency of 48.8%,47.2%,and 84.7% severally.It is verified that the proposed fusion planning algorithm has great advantages in both planning efficiency and path length.(3)Select a real unstructured road,and use the proposed algorithm to complete the path planning and tracking simulation test in this scene.An outdoor parking lot is selected as the simulation scene.Firstly,perform the threshold segmentation and geometric division.Then,according to the proposed planning method,successively complete the map processing,path planning using the improved algorithm,and redundant point removal operations.The results show that the path length is shortened by 4.8% and the planning time is shortened by 23%,and then the path is smoothed.Finally,model predictive control is used to track the planned path,and the maximum tracking deviation is only 4.4 cm.The experimental results show that the proposed improved graph search algorithm based on map processing has practicality and the applicability of MPC for tracking. |
PDF Full Text Request