Motion Planning For Autonomous Mobile Robots In Dense Crowds

With the development of robotic technologies,mobile robots are gradually deployed to social pedestrians environments,such as airports,hotels,and hospitals.Traditional robot motion planning approaches mainly focus on static or sparse pedestrian scenarios,without considering the impact of pedestrian behavior on robot motion.However,mobile robots moving safely and efficiently in dense crowds is a challenging problem because the interactive motions of pedestrians are increased and the free space for robot moving is reduced in dense crowds.To generate a robot motion,the motion planning framework needs to detect the surrounding pedestrians,and then predict their trajectories,and finally find a safe trajectory based on the predicted trajectories.This thesis seeks a motion planning framework to improve the safety,efficiency,legibility,and social compliance of mobile robot motion in dense crowds.In order to enable a mobile robot to understand dense crowds and to move safety,the problems of dynamic pedestrians detection and static environment cognition are studied.First,a lightweight pedestrian detection and tracking algorithm is proposed to achieve real-time pedestrian perception on the limited onboard computing resources of mobile robots.Second,a semantic segmentation algorithm is proposed to segment different regions and corresponding entrances and exits in the map.The robot can quantify the intentions of pedestrians by these entrances and exits on the map.In order to accurately predict pedestrians’trajectories,the problem of explicit mod-eling of pedestrians’intention and interaction is studied,and a pedestrian trajectory prediction method is proposed.First,the entrances and exits on the map are constructed as a set of pedestrian intentions.A naive Bayes model with a sliding window is proposed to predict pedestrian intention based on his/her historical trajectory.Then the proposed pedestrian trajectory prediction method is constructed by integrating the naive Bayes model into an optimal reciprocal collision avoidance model.In order to move in dense crowds safety,a trajectory planning method based on stochastic model predictive control is proposed.First,a composite objective function is proposed based on collision probability to weigh the safety and efficiency of candidate local trajectory.Thus the motion planning problem is formulated as a trajectory optimization problem.Then the optimal trajectory is calculated based on the stochastic model predictive control algorithm.In order to enable pedestrians to understand the robot moving behavior easily,and to improve the safety and efficiency of motion,a decision planning method of multi-modal-mode of the robot is proposed based on an explicit model of pedestrian walking decisions.First,based on the research of social psychology on pedestrian walking decision,three moving modes of the robot and corresponding trajectory sampling spaces are constructed.Thus the multi-modal-motion decision problem is formulated as a trajectory optimization problem.Then a composite objective function is proposed to evaluate the candidate trajectory considering safety,efficiency,and oscillation.In order to integrate the robot into pedestrians and improve the acceptance by human society,a motion planning framework is proposed based on an explicit model of collective and individual social conventions of pedestrian walking.First,a statistical model of preferred human flow direction based on mixed Von-Mises distribution is proposed to generate a human flow preferred-directional map.An extended hybrid A~*algorithm is proposed based on the human flow preferred-directional map to calculate the globally optimal path following the social conventions of collective walking.Then an optimal trajectory planning algorithm with self-adjusting control law parameters is proposed based on heuristic rules of social conventions of individual walking and metabolic energy objective function.Overall,based on robot motion planning challenges in dense crowds,this thesis proposes the new motion planning framework constructed by environmental perception,pedestrian trajectory prediction,and motion planning.Experiments show that the motion planning framework can integrate the mobile robot into the pedestrian flow and achieve safe,efficient,and socially-compliant motion in dense crowds.