| The proposal of the "Fourteenth Five-year Plan" promotes the continuous development of intelligent manufacturing in China.As the main equipment of material transportation,Automated Guided Vehicle(AGV)plays an important role in intelligent production and reducing the cost of material transportation.AGV is a complex and highly intelligent device,involving core technologies such as navigation,positioning and path tracking.These technologies are also difficult for AGV to get rid of the traditional working mode and become intelligent.The main work of this paper is to study the multi-sensor fusion positioning system of highly intelligent AGV,and design the lateral tracking control strategy of AGV for the desired path by using the positioning results.The core contribution of the whole research is mainly divided into the following four parts.Firstly,the sensors required by AGV positioning system are modeled and their internal and external parameters are calibrated.Especially for the internal and external parameters of camera-wheel tachometer,a calibration scheme based on TagSLAM framework is proposed.The principle of the calibration scheme is introduced in detail,and the rationality of the calibration algorithm is verified by experiments.Secondly,the framework of visual inertial navigation fusion positioning system is studied,including data preprocessing and visual inertial navigation joint initialization.Using wheel tachometer to solve IMU scale drift problem in visual inertial navigation fusion positioning system,the objective function of multi-sensor fusion positioning is constructed,and the Jacobian matrix of objective function to each variable to be optimized is derived.Finally,the visual inertial navigation positioning algorithm incorporated into wheel tachometer in this paper is compared with ORBSLAM3 algorithm in automatic driving data set KITTI to verify the accuracy of the multi-sensor fusion positioning algorithm designed in this paper.Thirdly,the path tracking control technology of AGV multi-sensor fusion positioning system is studied,and the second-order Runge-Kutta method is used to discretize the nonlinear prediction model directly to reduce the computational cost of solving the nonlinear prediction model.Aiming at the multiple iteration problem existing in the traditional model prediction algorithm when solving the desired path point,a lateral tracking control strategy based on local continuous path was proposed,and a nonlinear optimization problem based on local continuous path was constructed and solved by quadratic programming.Finally,the effectiveness of the lateral tracking algorithm was verified by simulation platform.Finally,the software and hardware experiment platform for multi-sensor fusion algorithm and path tracking algorithm is built,and the running effect of location tracking algorithm in real environment is verified on the platform. |
