| The robot’s autonomous operation in quasi-dynamic outdoor environments,such as campuses,parks,and factories,is essential to improve the intelligence of production and life.In this paper,to improve the autonomous navigation capability of robots in these environments,we make research on three fundamental aspects of robot navigation,localization and mapping,environment perception,and path planning,to address the problems in existing related research.Among them,simultaneous localization and mapping(SLAM)provides a priori map and localization information for robots to achieve path planning.Environment perception enables robots to sense dynamic people in these environments and provides accurate the locations and trajectories of them.Then,path planning algorithm plans paths to avoid dynamic and static obstacles based on localization information,generated maps,and person locations to make robots achieve navigation tasks safely and smoothly.These three components complement each other and work together to achieve robotic autonomous navigation tasks in these environments.The three main research and innovation points of this paper are shown as follows:At first,in terms of SLAM,this paper proposes a LiDAR SLAM algorithm based on environmental slope information for solving the problems of elevation error and map distortion in the practical application of existing algorithms.In this paper,the slope information in the environment is extracted to generate a slope factor,which is added to the back-end factor graph algorithm to optimize the robot poses.At the same time,an incremental angle estimation method is designed,which estimates the robot pitch angle by accumulating the slope information of the road surface where the mobile robot has passed,and realizes the real-time correction of the robot’s pose.By designing relevant experiments and comparing with other similar algorithms,the algorithm proposed in this study can effectively solve elevation errors and map distortions.Then,in terms of environment perception,this paper proposes a people detection and tracking algorithm based on sparse LiDAR point cloud data to cope with variable outdoor illumination conditions and to improve the accuracy of person recognition while ensuring real-time performance.In this study,a two pipeline convolutional people detection network is designed to identify people in the environment and determine their locations using extracted features of segmented point clusters.Then,the extracted people locations are tracked using the UKF algorithm to obtain the trajectories of the people in the environment.In comparison with similar existing algorithms,the proposed algorithm can locate the position of people in the environment more accurately while guaranteeing the real-time performance.In the public dataset L-CAS performance,compared with similar algorithms,it improves 30.5%in detection accuracy and reduces the predicted location error by 19.7%.At last,in terms of path planning,this paper proposes a hierarchical MPC-based path planning algorithm for solving the problem that the trajectory and velocity profiles are not smooth enough using the traditional MPC-based path planning algorithms.The proposed algorithm dynamically adjusts the prediction horizon of the MPC local path planner in the framework according to the crowding level in the dynamic environment to avoid the motion of people within the prediction horizon.The acceleration penalty term is also set in the loss function to achieve a smoother trajectory and velocity profile of mobile robots.In addition,an event-triggered strategy is used in local path planning to limit the prediction horizon of the designed MPC local path planner when the robot is in an open environment,which can reduce the time for the robot to calculate the optimal trajectory.In the experimental part,the proposed algorithm proposed can make mobile robots complete the navigation task more quickly and safely when compared with the traditional algorithm and the reinforcement learning-based algorithm. |
PDF Full Text Request