Research On Key Technologies Of Intelligent Agent Path Planning Under Multiple Operating Conditions

Artificial intelligence achieves the processing of complex problems by simulating human thinking processes,and its rapid technological development has driven the emergence and advancement of intelligent agents such as mobile robots,drones,and unmanned vehicles.These intelligent agents can be applied in various fields such as military,construction,and agriculture,etc.They are important tools for industrial automation and intelligence.Path planning is a crucial capability for these intelligent agents as it assists in navigation,obstacle avoidance,and target acquisition.Efficient path planning algorithms can reduce losses,ensure the safety of the agents,and help save time and system resources.The intelligent agent in this article refers to a multi-environment adaptive intelligent agent that needs to work in multiple environments,which places higher demands on its path planning capabilities and requires the simultaneous use of multiple path planning algorithms to select the most appropriate planning strategy for different environments and working conditions.The path planning of the multi-environment adaptive intelligent agent in this article is mainly divided into three working environments and working conditions: traditional path planning in 2D environment with map information,path planning in 3D environment with limited environmental information,and coverage path planning when multiple agents collaborate.Currently,path planning for intelligent agents can be achieved in some operating conditions.However,there are still many critical technological issues that need to be studied,especially when dealing with special working conditions or specific requirements for planning algorithms.Firstly,in traditional 2D path planning with known map information,the intelligent agents may waste computational resources during the exploration process when using existing reinforcement learning algorithms,leading to slow convergence.Secondly,in 3D path planning with limited environmental information,the local path planning of the agents requires complex and expensive sensors,and global path planning may be less efficient due to incomplete map information,and it tends to produce relatively conservative paths.Lastly,in coverage path planning for multi-agent collaboration,when faced with complex tasks or algorithms with many constraints,existing planning algorithms often produce non-smooth paths due to excessive number of corners,and the workload allocation among multiple agents may not be balanced,resulting in losses for the agents.Therefore,the main innovations of this article are as follows:(1)A new Paired Whale Optimization Q-learning Algorithm(PWOQLA)was proposed for path planning on traditional 2D grid maps.Firstly,a Paired Whale Optimization Algorithm(PWOA)was proposed to accelerate the search speed for prey.Secondly,a new Selective Exploration Strategy(SES)was introduced,which considers the relationship between the current position and the target position to reduce unnecessary exploration.Thirdly,a non-linear function was designed to dynamically change the value of ε in ε-greedy Q-learning based on the iteration count,focusing on exploration in the early stages and shifting towards exploitation in the later stages.Fourthly,utilizing the improved Whale Optimization Algorithm to initialize the values of the Q-table,the Q-table pre-learns from previous experiences before the exploration process,which can enhance the efficiency of the algorithm.PWOQLA was compared with other path planning algorithms,and experimental results showed that it achieved higher accuracy and faster convergence speed in 2D agent path planning.(2)A new Intelligent Agent-oriented Limited Environmental Information Path Planning Procedure(IEIPPP)was designed to address the issue of requiring extensive prior environmental information in current path planning methods.IEIPPP finds collision-free paths by analyzing a set of images of the planned area.Firstly,the image set of the planned area was processed using the COLMAP point cloud reconstruction tool to reconstruct a 3D point cloud model of the environment.Then,an improved Mechanical Selective Rapidly Exploring Random Tree Star algorithm(MSRRT*)was proposed for point cloud path planning.In MSRRT*,to reduce the high collision probability caused by excessive randomness,gravitation and repulsion forces were introduced to help correct the positions of random nodes.To stabilize the vertical fluctuations of the path,elastic potential energy calculation was introduced to balance the height differences between adjacent nodes.To reduce the computational complexity of the path planning algorithm,a target-based sampling strategy was proposed,allowing the algorithm to selectively sample with a certain probability.Finally,the effectiveness of IEIPPP was validated using different types of image datasets.The experimental results showed that IEIPPP can plan collision free paths with only image sets and without other sensor devices.(3)To address the problem of multi-agent coverage path planning,the Particle Swarm Optimization Improved Multi-robot Spanning Tree Coverage Star(PIMSTC*)algorithm and the Horizontal and Vertical Multi-robot Spanning Tree Coverage Star(HVMSTC*)algorithm were proposed.For weighted maps,a Spanning Tree Peer Reconnection Strategy(STPRS)was introduced first.Around the end nodes,according to the STPRS,it has been determined whether there are connection methods with smaller edge weights or fewer path turns to optimize the coverage paths.Subsequently,a Particle Swarm Optimization Cut(PSOcut)was proposed to optimize the partitioning strategy.Based on edge weights,the PSOcut optimizes the adjustment of the area size after uniform partitioning through the Particle Swarm Optimization algorithm,enabling multiple robots to work in a balanced manner and thereby improving the overall efficiency of the algorithm.Next,for unweighted maps,in order to reduce the number of turning angles in the planned path,the map information was analyzed before planning the path.The Horizontal and Vertical Segmentation Strategy(HVSS)was used to determine whether to segment horizontally or vertically.Subsequently,all segments were connected to form a spanning tree using the Segment Connection Strategy(SCS),and tasks were evenly distributed among multiple robots.Finally,experiments were conducted on simulated maps of different scales to validate the PIMSTC* and HVMSTC* algorithms.In summary,the research on key technologies for multiple operating conditions intelligent agent path planning can improve the planning efficiency,reduce planning time and turning angles in traditional 2D path planning when map information is available.In scenarios with limited environmental information,paths for intelligent agents can be planned after reconstructing 3D maps.It has achieved collaborative coverage path planning for multiple intelligent agents,optimizing the number of corners in coverage paths,and evenly distributing the workload among intelligent agents.