| In the past decade,with the development of mobile robots and artificial intelligence,mobile robots begin to enter the public society from the laboratory environment to provide public services for human.The scene of public service is more complicated,especially the pedestrian environment proposes a new challenge to the obstacle avoidance algorithm of mobile robots.Therefore,it is of great significance and promising to study the autonomous navigation algorithm of mobile robots in the pedestrian environment.The current autonomous navigation technology ensures that mobile robots operate in a regular environment such as a factory.However,in the pedestrian environment,the interaction between robots and pedestrians is faced with many theoretical and engineering difficulties.In order to cope with these challenges,this paper has carried out research and exploration from three aspects: obstacle avoidance,pedestrian perception and construction of mobile robot system.The specific contents and research results are as follows:1.The traditional obstacle avoidance algorithm can not meet the autonomous navigation requirements in the pedestrian environment.The improved dynamic window approach considers the pedestrian's state,but does not take into account the social norms and safety distance that pedestrians follow,resulting in robot and pedestrian interaction not friendly enough.Aiming to solve the problem,this paper proposes an obstacle avoidance algorithm based on deep reinforcement learning,which introduces the "right pass rule" and safety distance into the reward function design.Through the imitation learning and deep reinforcement learning,the training gets an obstacle avoidance strategy that maps state space to action space.Through the implementation of simulation experiments and physical experiments,it is proved that the obstacle avoidance algorithm based on deep reinforcement learning can effectively improve the interactive performance of robots and pedestrians.2.Pedestrian perception is an important component for robot to autonomous navigate in the pedestrian environment.Based on the advantages of visual semantic information and precise sensing distance of lidar,this paper designs a pedestrian detection and tracking system based on camera and 2D lidar.Firstly,the key points of the human body are detected by the improved human key point detection algorithm.Then the D-Means algorithm is used to cluster and track the point cloud processed by the map filter.Finally,based on the result of joint calibration,the probability fusion algorithm is used to fuse the detection results.The system obtains the trajectory of the pedestrians around the robot.3.This paper designs and implements an mobile robot test platform,including differential drive chassis based hardware system and ROS-based software system.The whole system is low cost and has great application prospects.For testing pedestrian detection and tracking,the algorithm combining lidar and camera is deployed on the robot platform.The experimental results show that the system can detect and track pedestrians in different poses,different sizes and different illumination conditions while the robot is stationary and moving.In the self-built data set for quantitative evaluation,the system designed in this paper can reach 89.2% detection rate when miss rate is 7.2% and false positive per minute is 1.1,which basically meets the sensing requirements of low-speed mobile robot.Finally,the obstacle avoidance experiment of mobile robot is conducted.The experimental results show that the obstacle avoidance algorithm designed in this paper can be successfully deployed from the simulation environment to the physical environment,and the autonomous navigation task can be completed under the natural interaction and pedestrian intentional blocking scenarios.The obstacle avoidance algorithm follows the "right pass rule" when interacting with pedestrians.The experiment verifies the effectiveness of the obstacle avoidance algorithm in the pedestrian environment. |
PDF Full Text Request