With the implementation of "made in China 2025" strategy and the vigorous development of logistics and storage industry in unmanned chemical plants in recent years, the market demand of automated guided vehicles (AGV) has increased sharply. In the modern manufacturing and processing system, the logistics transmission time accounts for 80% - 90% of the total production time of the whole product, and the transmission and storage costs of production materials account for 30% - 40% of the total parts processing cost [1-2]. Shortening the transmission time of AGV in logistics plays an important role in reducing the cost of transmission and storage in the whole logistics process. Since omnidirectional AGV is used in the factory and warehouse, it is necessary to ensure that AGV car can travel linearly between each station. Usually, the navigation function package in the ROS system of the upper computer completes the optimal path planning based on the AGV car's position to the target point. The linear path is resolved into a series of speed instructions, which are received and controlled by the lower computer in the AGV At the same time, the chassis transmits the attitude data collected by the attitude sensor to form a closed-loop attitude control. In order to realize the AGV linear driving based on slam navigation, the problem of attitude calculation based on omni-directional AGV must be studied to obtain the accurate attitude calculation data of AGV chassis and complete the attitude control of the whole car.
The attitude of AGV is the most important part of sensor information for slam mapping and automatic navigation. The accuracy of attitude data directly affects the accuracy of attitude control and positioning. The parameters of attitude description mainly include cosine matrix, Euler angle and attitude quaternion . Attitude quaternion has the advantages of small amount of calculation, fast calculation, full attitude, and can avoid the "universal joint deadlock problem", which has been widely used in attitude control system. But quaternion belongs to rotation vector, and the four variables are not independent. In reference , quaternion is used to fuse with Kalman filter algorithm, but the system is linear. The measurement noise and system noise are introduced into the state equation to improve the attitude solution accuracy, but errors are introduced when observing the linear equation. In reference  and , quaternion is used in unscented Kalman filter algorithm, which solves the error caused by linearization, and has higher accuracy than extended Kalman filter algorithm. In reference , a complementary filtering algorithm based on the fusion of accelerometer and gyroscope is adopted. According to the external force detected by accelerometer, an adaptive algorithm is designed to modify the fusion weight, which reduces the measurement inaccuracy caused by the nonlinearity of the system to a certain extent. However, due to the integral drift of gyroscope, the yaw angle always exists drift. In reference , after the geomagnetic field in the navigation system is rotated to the carrier system and cross multiplied with the three-axis magnetometer, the error vector is obtained to correct the gyro drift. However, since the magnetometer is directly involved in the quaternion updating, the attitude angle is easily disturbed by the magnetic field, and a new interference error is introduced into the estimation of the horizontal attitude angle.
Based on the deficiency of the above research, aiming at the integral drift error of gyroscope, this paper combines the information of gyroscope and accelerometer based on the quaternion updating equation of horizontal attitude angle, optimizes it by conjugate gradient correction method, and estimates the horizontal attitude angle in real time. Under the condition that magnetometer is easy to be disturbed, the attitude of the yaw angle of AGV is calculated separately by using the first-order weighted complementary filter, so that the yaw angle is not involved in quaternion calculation, so as to avoid the horizontal attitude angle calculation error caused by magnetometer.
1. Kinematics model and attitude description of AGV
In this paper, a four-wheel independently driven Mecanum AGV car based on laser slam navigation is adopted, which has high flexibility and can realize omni-directional movement in any plane. Micro electro mechanical system is used to realize attitude calculation, and the attitude description of AGV car is carried out [9-10].
1.1 kinematics analysis of AGV
The omni-directional AGV chassis adopts the Mecanum wheels which are symmetrically distributed on the left and right wheels. At the same time, the axial direction of the rollers of the Mecanum wheel on the diagonal line of the chassis is the same. The different steering of the four wheels is controlled by the embedded main control board at the bottom layer, so as to realize the omnidirectional movement of the AGV car in the plane . The omni-directional AGV adopts four-wheel drive of DC motor, has three degrees of freedom on the plane, has good maneuverability and the acceleration performance is relatively uniform in all directions .
Please download the rest