On navigation of a robotic land vehicle

Abstract

Estimation of the state of a robotic vehicle with independently-driven wheels is considered. The sensors in this application measure wheel rotation, angular velocity, and range from a fixed station. In addition to use of an extended Kalrnan filter for fusion of the data from these sensors, a simplified technique to reduced computational requirements is presented. Using dynamic parameters of an existing experimental vehicle, performance of the filters was simulated and compared. It was found that the simplified technique rivals the extended Kalman filter in performance but has a significant computational advantage. Introduction This paper is concerned with one aspect of navigation of a robotic vehicle, namely the estimation of the vehicle's state. The type of vehicle considered has two independent wheels, each driven its own motor. Such vehicles are popular for robotic applications for several reasons: They are capable of zero-radius turns and their dynamics are relatively simple. The complement of sensors to be considered includes the following: 0 Incremental shaft encoders to measure the angular rotation of each wheel. 0 A rate gyro to measure the angular velocity of the vehicle. 0 An ultrasonic range sensor that can measure the distance (range) from a transmitter on the vehicle to a receiver fixed with respect to the coordinate system. In principle, the encoders can determine the change in angular orientation of the vehicle, so the rate gyro is not essential to the navigation system. Owing to wheel slippage and possible encoder miscounts, however, the rate gyro enhances performance of the system. The combination of encoders and the rate gyro are not sufficient, however, to determine absolute azimuth 'This investigation was supported in part through a grant (Project No. 994803) from the Center for Manufacturing Systems at NJIT 2~istinguished Professor, Electrical and Computer Engineering and Center for Manufacturing Systems. Associate Fellow, AlAA 3~raduate Student, Electrical and Computer Engineering RECOMR Figure 1: Robotic vehicle with independently driven wheels. of the vehicle: The readings of all the sensors are independent of the initial orientation of the vehicle. A single (error-free) range measurement, however, is sufficient to resolve the orientation uncertainty. Hence the presence of the ultrasonic range sensor results in a system that is completely observable, intuitively and in the formal nonlinear sense [I]. Merging of the data from all the sensors is readily accomplished by use of the well-known extended Kalman filter (EKF) algorithm in which the EKF gains are calculated by numerical propagation of the corresponding matrix Riccati equation, a substantial computational burden. XI reduce the burden a suboptimal hierarchical approach is considered in which the attitude and velocity is obtained by using a steadystate, fixed-gain Kalman filter and these estimates are used to drive a second-stage estimator which uses the range measurement provided by the ultrasonic sensor.