Research On Path Planning Technology Of Mobile Robot In Indoor Complex Environment

Autonomous navigation is an important prerequisite for mobile robot to achieve intelligence,and path planning technology is one of the key technologies for autonomous navigation of mobile robot.With the increasing complexity of mobile robot application scenarios and obstacles tend to be high dimensionality in the environment,it is necessary to study an algorithm that can achieve real-time obstacle avoidance and plan a high-quality path in dynamic complex environment.The path planning and obstacle avoidance technology of mobile robots in indoor complex environments with high-dimensional and dynamic obstacles were studied in this paper.Build an embedded mobile robot platform based on ROS system and adopt camera and 3D lidar in dynamic obstacle avoidance scheme.An path planning framework was proposed which combines an improved A * global path planning algorithm with improved reinforcement learning local path planning algorithm.The effectiveness and advancedness of the proposed method were verified in simulation environment and actual environment respectively.The following works have been finished in this paper:First,the advantages and disadvantages of existing environmental perceptive sensors were analyzed in mobile robot obstacle avoidance,and an obstacle avoidance scheme combining binocular camera and 3D lidar was proposed to realize the perception of high-dimensional obstacles in the indoor environment.Second,the advantages and disadvantages of several commonly used global path planning algorithms were analyzed.The traditional A * algorithm has many shortcomings in path planning in large-scale environment,such as poor real-time planning path,many path turning points and difficulty in controlling turning angles.An improved A * algorithm was proposed,in which the cost of the parent grid is introduced into the heuristic cost,the weight is introduced into the cost function,and the redundant path points and turning points are eliminated by smoothing the original path.Compared with different global path planning algorithms in simulation environment to verify the superiority of the improved A * algorithm proposed in this paper.Then,the application of reinforcement learning algorithm in path planning was studied in the local path planning algorithm.The traditional Q-learning algorithm has the problems of slow convergence speed and poor path smoothness in path planning.An improved Q-learning algorithm was proposed,in which the attractive potential field in the artificial potential field method is introduced in the process of Q value initialization,increase the step size and adjust the direction of behavior.The effect of different improvements on the performance of the Qlearning algorithm was compared in the simulation experiment,and the superiority of the improved Q-learning algorithm proposed was verified.Finally,a prototype platform based on ROS was built and the path planning algorithm proposed was transplanted into navigation package.Some prototype experiments have been completed in static environment,high-dimensional obstacle environment,dynamic environment and complex environment respectively.The experimental results demonstrate that the superiority of the proposed path planning algorithm and the effectiveness of the obstacle avoidance strategy in a high-dimensional dynamic environment...