Mobile Robot Navigation Based On Situational Experience And Sparse Point Cloud

Autonomous navigation,which is the core capability of mobile robot system,integrates localization,mapping,path planning and control.The mapping process realizes the understanding of environment.Sparse point cloud map can sparsely reconstruct the environment in real time.However,the sparse point cloud cannot fully express the environment information,so it is difficult to use it independently to complete the navigation tasks.Human beings can realize experience-based navigation ways by learning the situational information of the environment,rather than relying on high-precision environmental maps.This paper proposes a novel mobile robot navigation system which realizes environmental cognition based on situational experience and sparse point cloud.The framework of the navigation system consists of three modules: sparse point cloud,situational experience map and path planning.The mobile robot localizes itself based on sparse point cloud,and plans the global event sequence and local path with the situational experience map.Firstly,the sparse point cloud of environment is reconstructed and a multi-sensor fusion locating scheme is constructed.The system of building sparse point cloud adopts the feature-based fronted algorithm to track pose,and uses the optimized-based backend algorithm to update map.The sparse point cloud,which is used to locate robots and assist in constructing situational experience map,can be updated online according to the environmental changes.In order to achieve more accurate and robust navigation performance,a fusion locating scheme of visual odometer,IMU and wheel odometer based on Extended Kalman Filtering is constructed by the complementary analysis of multi sensor.Secondly,the situational experience map is constructed.Inspired by human navigation ways with situational experience,the model of situational experience constructed encapsulates scene perception,posture,event transfer set and decision behavior.This paper proposes the situational incremental learning method and self-organizing cognitive experience to construct the situational experience map,which forms a comprehensive description of the geometric and topological relationships of the environment.The situational experience map simplifies and simulates the experience accumulation process to complete the environmental cognition,which provides support for the path planning and navigation of mobile robot.Then,path planning algorithm is designed with constructed situational experience map.The planning algorithm of global event sequence is designed by combing the event transfer set of situational experience with Dijkstra algorithm to obtain the global optimal path.Improving the evaluation function of dynamic window approach with the decision behavior of situational experience,the e-DWA algorithm is proposed to obtain the optimal path in the local range.Finally,the mobile robot hardware platform and software system is built and navigation experiment is designed.The hardware platform is HUSKY A200 mobile robot equipped with Kinect2 camera,IMU and laser sensor Hokuyo UTM-30 LX.Based on the kitti data set and the actual indoor scene,the proposed navigation system was used to perform the mapping and navigation experiments.The experimental results show this navigation system can generate the global optimal path according to different navigation tasks with high navigation accuracy.