Research On Sampling-Based Motion Planning Algorithm For Robots In Dynamic Scene

At present,the research and application of mobile service robots are receiving more and more attention.Compared with industrial robots,the close interaction of mobile service robots with people proposes a challenge to robot navigation.This means that the robot not only needs to navigate in the dynamic scene,but also considers the movement of the pedestrian and interacts with the user in a humanized way.Based on multi-sensor estimation of pedestrian pose and motion information,this paper realizes the intelligent humanized path planning method for mobile robots,and realizes robot navigation in the environment where humans and robot coexist.Firstly,this paper builds a robot platform based on the requirements of robot navigation in the dynamic environment.Then we test the robot platform and build a kinematic model for it.In this paper,the gmapping algorithm is used to construct and locate the indoor environment.We research the positioning and relocation performance of the gmapping algorithm.By optimizing the number of sampled particles,a static map with accuracy matching navigation requirements is established.Secondly,this paper studies the rapidly-exploring random tree algorithm,and proposes three path optimization methods for the shortcomings of the algorithm.Aiming at the efficiency problem of RRT algorithm search path,the influence of nonsynchronization on search efficiency is analyzed,and the appropriate step size is selected according to indoor scene.For the path non-optimality generated by RRT algorithm,a pruning algorithm based on distance optimization is proposed,which reduces the cost of the path generated by the RRT algorithm.With using the cubic Bezier curve for smoothing,a feasible path is obtained to meet the kinematic parameters of the robot.Thirdly,pedestrian recognition and motion prediction based on multi-sensor fusion were studied.The mobile robot is designed to coexist with users in the workspace and needs to identify the target user and its intent in real time.In this paper,we use the Lidar and the Kinect together for human detection.We performed calibration between the laser and the camera.The depth information template was used to match the upper body position of the human body.The laser data was used to identify the human leg information.According to the scalar optimal weighted fusion method,the laser and camera data fusion was performed.The skeleton gains pedestrian orientation and improves the accuracy of pedestrian detection.Finally,the robot navigation system based on pedestrian perception is studied.Pedestrian information is added to the multi-level cost map by establishing a personal space model.Then we propose an improved algorithm based on repaidly-exploring random tree which named Risk-DRRT*.This algorithm can perform local path planning in a dynamic environment.The Risk-DRRT* algorithm inherits the probability completeness of the RRT algorithm and takes into account the comfort of pedestrians which enabling human-machine-friendly navigation in dynamic environment.In summary,the proposed algorithm is experimental in the simulation and real world.The humanized mobile robot navigation system which we proposed provides satisfactory performance in a dynamic indoor environment.