Autonomous Navigation For Mobile Manipulation Robot In Human-Robot Coexisting Elevator Environment

With the development of urbanization in China,the number of permanent residents in cities and towns has increased rapidly.Buildings have become an important space for more urban residents to live and work.Service robots have broad applications in hotels,office buildings,hospitals and other buildings,it is necessary to navigate on multiple floors.When a service robot takes multi-floor tasks in a building,taking an elevator is an inevitable and important process.This thesis focuses on the robot only relying on its own sensors and manipulator to operate and take the elevator without modifying elevator devices,realize the robot navigating to elevator considering efficiency,pedestrian motion and interaction rules in humanrobot coexisting environment.The navigation of the mobile manipulation robot in human-robot coexisting environment can be modeled as Markov decision process.This thesis designs a Markov decision model that conforms to the elevator scene by analyzing a large amount of data of the real environment.By collecting a large number of pedestrian trajectories in real scenes,this thesis analyzes data from four dimensions of velocity,time,space,and social rules to obtain the required parameters in the Markov decision model.Through the statistical analysis of the data obtained by human remote controlling robot,it is determined that the movement way of robot,the suitable velocity of robot and the polite distance between human and robot.The constant velocity prediction model is the most suitable model by trading off computation efficiency and prediction accuracy of four pedestrian motion trajectory prediction models.Thus,the constant velocity prediction model is incorporated into Markov decision model to predict trajectories.The interaction rule of “first out,last in” is incorporated into the proposed Markov decision model through a precondition function.Through the constant velocity prediction model,the future trajectory of pedestrians that cannot be directly observed is incorporated into the pedestrian state.A reward function is designed according to efficiency and politeness required by the task.According to the results of data analysis,the model is parameterized and the parameters and indicators of the Markov decision model are determined.The experiments show the proposed navigation algorithm can trade off efficiency and the comfort of pedestrians.Finally,the operation plan and the operation planning scene are optimized for the problems existing in the operation module.The feasibility of the improvement of the operation plan is verified through the autonomous navigation and operation in elevator environment.Compared with the earlier work of the research group,the time indicators of each task module of the improved experiment are analyzed.The time of the whole process of autonomous navigation and operation in elevator environment has 24% improvements.