| Intelligent indoor service mobile robots are becoming the focus of the robotics industry.When discussing whether mobile robots can solve practical problems to achieve autonomous movement,accurate positioning is the core key technology to achieve autonomous navigation of mobile robots.At present,most mobile robots have a single application scenario,which makes it difficult to cope with the complex and changing environment during actual use.How to make mobile robots efficient and accurate in the face of different positioning problems is the first step towards intelligent mobile robots.In this paper,we investigate the localisation of mobile robots in indoor environments by fusing optical flow sensors and Ultra Wide Band(UWB)sensors with mainstream sensor fusion solutions to improve the adaptability of mobile robots to different scenarios as well as the efficiency,success rate and accuracy of robot localisation.First,for the relative robot localization problem,in addition to the conventionally used wheeled wheel odometer localization based on the two-wheel drive model and LIDAR odometer localization based on the Iterative Closest Point(ICP)algorithm,this paper proposes a rotation-compensated optical flow sensor odometer method based on the optical flow sensor(originally used for UAV fixed-point hovering)This paper proposes a rotationcompensated odometry method based on the optical flow sensor(originally used for UAV fixed-point hovering),which is applied to the positioning of a mobile robot.The rotation of the optical flow sensor is compensated for by fusing the Inertial Measurement Unit(IMU).Finally,a multi-stage Kalman filter is used to fuse the information,with the optical flow sensor positioning and wheeled odometer positioning providing the high frequency of the fused position estimation,the LIDAR providing the accuracy of the fused position estimation,and the fused position estimation resetting to correct for the larger optical flow sensor and wheeled odometer errors.Then,for the abducted robot problem,the wheeled odometer and LIDAR cannot perform the position estimation properly under normal conditions,this paper uses the optical flow sensor to build a motion model based on the optical flow sensor to obtain a rough relative position estimate,and uses a conditional variational self-encoder generation model to train the relative position estimate output from the motion model to improve the relative position estimate based on the optical flow sensor.This improves the accuracy of local localization during the "abduction" process,allowing the robot position to be updated in the vicinity of the exact pose,thus enabling the robot to recover its position quickly and basically solving the problem of abducting the robot in a short time over a short distance.Secondly,for the absolute robot localization problem,the Monte Carlo localization algorithm based on particle filtering can basically solve the local and global localization problem,but cannot solve the abducted robot problem,and reduce the probability of losing the correct poses by adding random particles to improve the robustness of the algorithm.The efficiency of the algorithm is then improved by adjusting the size of the sample set through the KLD sampling algorithm.An adaptive Monte Carlo localisation algorithm that combines the addition of random particles and the KLD sampling algorithm is given.On top of this algorithm,UWB sensors are added to aid localisation to improve the probability of successful localisation and the efficiency of the algorithm.Based on the different noise parameters of the UWB sensors themselves for different obstacles,an occupied raster map with obstacle noise is proposed to further improve the adaptive Monte Carlo global localisation algorithm by converting the initialisation process from random particle generation for the whole map to particle generation for the possible map regions,which greatly reduces the probability of robot localisation failure and substantially improves the algorithm efficiency.Finally,to verify the feasibility and effectiveness of the algorithm.In this paper,a mobile robot experimental platform is built to validate the proposed robot localisation algorithm for different localisation problems in a real-world environment.The performance of the robot localisation algorithm is analysed and evaluated in terms of several dimensions,including the robot localisation trajectory(i.e.accuracy),the timeliness of the algorithm and the success rate of the algorithm in retrieving the localisation.The experimental results demonstrate that the relative positioning method proposed in this paper is characterised by high real-time performance,accurate positioning accuracy and good robustness;the proposed absolute positioning algorithm significantly improves the probability of robot retrieval and the efficiency of the algorithm,and effectively assists in solving the abduction robot problem;the proposed relative positional output based on the conditional variational self-encoder trained optical flow sensor motion model effectively improves the relative positional error from the real positional The proposed conditional variational self-encoder-based training optical flow sensor motion model output relative poses effectively improves the error with the real poses,basically solving the abducted robot problem in a short distance and short time. |
