| The application of mobile robot in human life is more and more extensive.The premise of realizing the robot to work independently in the environment is that it can complete the autonomous navigation task.The navigation of mobile robot is divided into map building and path planning.Map building is the process of robot cognition environment,and path planning is a necessary condition for robot to quickly perform tasks in the environment.Humans and animals also have environmental cognition and navigation abilities.These abilities are closely related to the spatial cognitive mechanism of brain.From the perspective of brain-like navigation,this paper proposes a bionic navigation method of accurately constructing experience map and quickly planning path.The hippocampus-medial entorhinal cortex region is the main part of the navigation system in brain,which contains many localization cells related to spatial cognition.In this paper,the head direction cell model is established by one-dimensional continuous attractor,and the angular velocity information of mobile robot is integrated to form its angular representation.The grid cell model based on phase coding is used to integrate the linear velocity information of mobile robot to form the grid coding of multi-dimensional representation in the current place.Based on the one-shot learning rule,the connection weights between grid cells and place cells are established.The place cell is activated by grid coding to form the place representation of robot.Through the cells model system,the pose of the robot is accurately expressed.The external environment information of the robot is of great significance to the building of the experience map and can provide a wealth of reference for robotic navigation.Based on the image information obtained by the visual sensor,the landmark information existing in the environment is identified by using the yolov4 algorithm,and the ORB feature points are detected to form the image features.The head direction cell information,place cell information,landmark information and image features are integrated into experience node.A complete experience map is constructed through the connection between experience nodes.In the process of building the experience map,loop detection algorithm is used to ensure the consistency of the experience map to build an accurate map.The experience map is the result of robotic cognitive environment,and it is also the collection of robotic state in the environment.Based on the sequence decision function of the hippocampus,the successor representation model of robot’s state is constructed through the reinforcement learning method.Combined with the goal-based reward function,the predictive map is established,and then the shortest global path to the target point is quickly planned.On the basis of the global path,the dynamic window algorithm is used to plan the motion trajectory conforming to the kinematics model,so that the robot can reach the target point stably.In addition,building a search tree based on landmark information enables the robot to quickly localize in the environment before performing navigation tasks.Experiments are carried out in simulated maze,KITTI data set and indoor environment respectively.Experimental results show that,compared with other bionic navigation,the proposed method has shorter calculation time and higher accuracy.Compared with the visual navigation method,it can achieve robust map construction without complex system.The experiment shows that the navigation method based on predictive map can adapt to the change of environment structure and plan the shortest path of current environment state. |
PDF Full Text Request