Research On Autonomous Collision Avoidance And Learning Method For Mobile Robot

Mobile robot works in unknown environment will encounter some obstacles,the ability of autonomous avoidance in a short time is the premise and foundation to ensure its safety to complete its mission.Usually,robots use their own sensors to detect environmental information for path planning.The information includes known environmental information and unknown environmental information.In the unknown environment,there are both static and dynamic obstacles,different types of obstacles have different methods to avoid collision.The learning ability of robot is one of the important indexes to reflect its intelligence level.At present,deep learning algorithms have been applied to many aspects of life,especially in the field of unmanned vehicles.Many unmanned vehicles have reached a practical level.In this paper,a deep learning algorithm is designed for mobile robot to avoid collision.The algorithm takes environmental information as input and the teacher system's result as a label.By the contrast between training results and labels,the algorithm can improve network parameters and enhance environmental adaptability.This paper can be divided into three parts:First of all,the paper analyzes the research status of mobile robots' autonomous avoidance behavior and establishs the robot model and coordinate system according to the characteristics of the the environment.Secondly,based on the existing information and the information detected by the sensor,an ant colony algorithm is designed to avoid the static obstacle.In addition,an improved artificial potential field method is used as an emergency avoidance algorithm to increase the safety of collision avoidance.Some collision avoidance strategy for dynamic obstacles in an unknown environment is also designed so that the robot can successfully avoid dynamic obstacles.The above method is verified by simulation and the physical verification is carried out by using the "Traveler No.4" mobile robot.Finally,the paper analyzes the principle of deep learning.A GRU-RNN network structure is designed for the path planning of mobile robot.At last,the GRU-RNN is trained and we can verfy its effection.