Mobile Robot Episodic Cognitive Map Building And Behavior Planning

Map building,autonomous locating and target navigation are the main functions and key technologies for autonomous mobile robots,which are also the hot and difficult points in the field of mobile robot.Physiology research shows that hippocampus-entorhinal cortex in the biological brain is the core area of spatial cognition.It can realize self-location through cells of space positioning system such as place cells and grid cells,which plays an important role in spatial navigation.Experience memory with the environment is formed through imitating the information processing mechanism of mammal's positioning system cells and combined with episodic memory.And the problem of the environmental cognition and behavior planning of mobile robot are studied.According to the background of mammalian spatial cognition,a spatial cognition model of mobile robot is constructed.The hippocampus-entorhinal cortex positioning system cells model is constructed.The path integration for velocity information has been accomplished by using the grid cells.And unimodal place cells firing field has been constructed through the iteration of grid cells with multiple layers by Hebbian learning method,thus expressing the space location of the robot by using the relative location of the place cells firing field.Based on the formation of episodic memory,state neurons is used to imitate place cells,and events model which encapsulated scene perception,state neuron and pose perception is established.An episodic memory model is constructed with events sequence,which is used to store the environmental perception information of the mobile robot,and realizes the generalized modeling of the environment.For the problem of the mobile robot's learning and exploration to the unstructured dynamic environment,a method of episodic cognitive map building is proposed,by imitating the information processing method of positioning system cells and organizing method of episodic memory in the hippocampal structure.The topological description of actual environment in mobile robots is realized.According to neuron activating mechanism,using scene perception to active different neurons to express different states of the robot,and generating state neuron sequence to represent the robot state set.And place cell model is used to output position information so as to coordinating the events.Meanwhile,loop closure detection is implemented by using perception information of color image,and an episodic cognitive map is constructed finally.According to the neuron sequence information stored in the episodic cognitive map,events reconfiguration is accomplished by utilizing state neuron positioning in memory space,and optimal global planning strategy of neuron set sequence is presented.When facing multiple paths to choose,the mobile robot can predict an optimal global planning path on the global scale according to episodic experience.Then calculating the adjacent events pose changes through pose perception in episodic memory.And a robot behavior control algorithm based on linear velocity and angular velocity control is proposed,which realized the task that the robot arrive at the target point along the planned path.Experimental platform composed by Husky A200 mobile robot and Kinect2.0 vision sensor is established.Based on robot operating system(ROS)and MATLAB software,cognitive learning and behavior planning experiments are implemented respectively in the typical environment,such as offices and corridors.And the comparison is made with the RatSLAM algorithm.Experimental result shows that the algorithm mentioned in the paper can achieve episodic cognitive map building in the unstructured dynamic environment,and accomplish the optimal behavior planning task based on the map.