Design And Implementation Of Human-aware Navigation System For Service Robot

Human-aware navigation is one of the core technologies in the field of service robot research.The combination of pedestrian modeling and path planning is an important trend in the current development of human-aware navigation for indoor service robots.According to the working environment of indoor service robots,a framework of human-aware navigation system for service robots with good scalability and stability is proposed,which conducts wide research on related technologies and designs software for each module of the system.The related research of human-aware navigation technology for service robots is summarized and the requirement analysis is carried out based on the technical difficulties in system design and implementation.A framework of human-aware navigation system for service robots with good scalability and stability is proposed based on the robot operating system ROS and discussed in detail.Firstly,considering the demand of pedestrian perception and modeling in the human-robot integration environment,the pedestrian tracking module is developed based on a detection-based tracking architecture,which implements multi-modal and multi-view pedestrian detection and tracking.Furthermore,a voting function is improved for the existing group recognition method based on F-formation and a consistent motion feature is introduced for dynamic group recognition.Base on the above pedestrian perception,the personal comfort distance and group interaction of pedestrians in the environment are modeled as social costs.Secondly,considering the demand of the path planning in the human-robot integration environment,the disadvantages of traditional path planning methods are discussed in detail and then an improved path planner is proposed.Multi-layer dynamic cost maps are generated containing the social cost for motion and group information of global pedestrians at different time-steps based on pedestrian trajectory prediction,which provides social constraints for global path planning.The proposed path planner builds a search-graph based on the defined lattice state and performs heuristic searches for the optimal path with a heuristic cost function based on the multi-layer dynamic cost-maps.In order to improve the adaptability to dynamic working environments,the ‘plan-prediction-execution' cycle is introduced for dynamic path search and motion control.Thirdly,considering the demand of the user-oriented semantic navigation,the task management and control module is designed to manage the semantic navigation subtasks.In order to improve the intelligence of navigation,modules of pedestrian retrieval and approaching pose calculation are developed for the identity and the approaching pose for the target user.Fourthly,considering the demand of task interaction and system monitoring,a human interaction interface is designed and implemented based on the rosbridge Web APIs.Finally,combined with the working environment of the service robot,functional and comparative tests are performed on each module of the system.A demonstration of the overall operation and abnormal situation handling for path planning and pedestrian retrieval in the delivery task scenario is given to verify the effectiveness and stability of the entire system architecture and each functional module.