Research On Mobile Robot Indoor Environment Exploration And Path Planning

Environment exploration and path planning is the basement of the mobile robots to autonomously perform the tasks in the indoor environment.The exploration whose purpose is to build the environment map,means the robot can efficiently carry out environmental traversal,and obtain the appropriate environmental information, so as to establish the environmental topologic map that the robot can understand in its brain, and complete the global path planning on the basis of the map, and ultimately achieve the completion of its tasks.In this paper, the researchment is focused on the environment exploration and path planning, the main study work of this paper is as follows:1. Mobile robot needs to efficiently complete the environment traversal to obtain environmental information as much as possible.However, the traditional random search method is inefficiency and waste of resources in some aspects,therefore,we proposed the traversal algorithm based on the finite state automaton FSAM(or finite-stateautomachine) for the mobile robot to explore the indoor environment.Firstly,the environment information in the robot probe domain will be rasterized,and the we designed different switching states,and studied the state transition strategy,desigend three state transition strategies: maximum entropy selection strategy, relocation grid selection strategy, movement search strategy. FSAM will organized the states and the state transition strategies into a whole loop.Finally,the robot can achieve the goal of efficient environment traversal.2. Based on the ensurement of the successful environmental traversal in the last chapter,aiming to build the topological map of the environment, a kind of variable threshold dynamic growing self-organizing neural network(Variable Threshold Dynamic growing Self-organizing Map, VTDGSOM) was proposed to build the environment model. the introduction of hierarchical clustering factor(hierarchical clustering factor, CF), allows network growth threshold to change adaptively according to the environmental information,and so as to achieve different levels of clustering,finally shorten the time of building environment model.In three typical environment for algorithm simulation, we compared the VTDGSOM algorithm with the existing SOM algorithm and DGSOM,the experimental results show that the VTDGSOM algorithm is more effective and efficient in the use of environment topological map creation.3. On the basis of the environment map which has been created in the last chpater, we proposed an algorithm combined pruning algorithm and A* algorithm(Pruning Algorithm_A*, PA_A*) for global path planning.PA_A*algorithm will do further cut optimization on the optimal path which is searched by A* algorithm.The simulation results show that the number of waypoints decreased and the distance of the path was shorten,and the efficiency of the robot global path planning was further improved.4. In the reinforcement learning of robot navigation,it may occur "curse of dimensionality" and learning performance problems due to the expansion of the state space. the proposed model combined BP neural network with Q learning. Designed inspired return function-Shaping function which is generated by BP neural network,and Shaping function was used to inspired update the Q learning lookup table. The generalization and prediction ability of neural network was combined with Q learning for robot indoor navigation.The experiments show that the method has a good effect on Q learning.The learning performance has been improved.