Research On Position And Attitude Perception For Self-driving Roller

The technology of unmanned roller,which is highly expected to improve work efficiency and accuracy,has attracted wide attention to take the place of manual operation in the construction industry.However,the accuracy of automatic operation is affected by the complex environment and the articulated structure of two frames,making its control challenging.One of the difficulties in solving this problem is the accurate perception of position,heading angle,pitch angle and roll angle.In this paper,in order to design more complete perception algorithm,based on the existing sensor sensing system structure of the roller,the estimation algorithm of heading angle and attitude is proposed for redundancy,and the fusion algorithm of position is proposed for precision.First of all,in order to design a redundant and accurate perception algorithm of position and attitude,a measurement platform is built according to the articulated mechanism of the roller and the needs of the sensing task,combining a global position system(GPS),an articulated angle sensor and an attitude and head reference system((AHRS))installed on the roller.In order to have a clear understanding of the characteristics of the sensor,measurement models of AHRS(including gyroscope and accelerometer)and articulated angle sensors were constructed.Secondly,in order to design the redundant output of the heading angle,a heading angle estimation model based on data-driven modeling is constructed.Empirical equation of heading angle estimation is established by analyzing the principle of AHRS and output data of AHRS.After that a heading angle estimation model is constructed by calibrating variables in the equation.Experiments show that the maximum error between the output of the heading angle estimation algorithm and the output of the GPS heading angle is 0.45 °,and the fitting effect is good.Finally,aiming at the problem of whether the GPS output heading angle and the estimated heading angle are credible in the operation of the roller,a judgment and optimization model based on the heading angle kinematics model for threshold prediction is constructed.Thirdly,in order to design the redundant output of the pitch angle and roll angle,the attitude estimation model of the inertial sensor assisted by the equation of motion is established.The state equations based on the gyroscope measurement model are constructed.And the measurement update equations based on the accelerometer measurement model,statistical model,and rotational motion model are constructed.Experiments show that the average error between the roll angle estimation value and the sensor measurement is 1.2 °,and the average error between the pitch angle estimation value and the sensor measurement is 0.45 °.Finally,an accurate positioning algorithm based on loosely coupled sensors and neural network is proposed.The time update equation based on accelerometer and gyroscope and the measurement update equation based on GPS and kinematic constraints are constructed.A loosely coupled algorithm based on error state extended Kalman filter is also designed to fuse the eastward position information and northward position information.The experiment proves that the loose coupling system can better compensate the position information when the positioning antenna is shaken to cause positioning deviation,and the GPS is in short-term failure(within 3s).However,loosely coupled algorithms are affected by environmental errors,resulting in excessive system errors.We propose a joint algorithm of reverse neural network and loosely coupled algorithm.Back-propagation network is established to compensate the loosely coupled algorithm.Simulation experiment analysis: after offline training and online learning of the BP neural network,the accuracy of the position information output by the time update process in the loosely coupled algorithm can be improved.In summary,the estimation algorithms of heading angle and attitude are proposed for redundancy,and the fusion algorithm of position is proposed for precision.They can effectively solve the problems of perception,including the insufficient diversity and the insufficient resistance.The algorithms proposed in this paper lay an important foundation for designing a complete perception algorithm.