Solutions for Tractor-trailer Path Following at Low Speed

Abstract

Autonomous tractor-trailer systems have become more and more popular in real life. This dissertation focuses on control and estimation for the path following problem where space-based path is used as reference. The goal is to make the trailer path follow the reference path as closely as possible. A kinematic model is derived and is applicable when the tractor velocity is relatively low. To reduce tracking error near the intersection between two different kinds of reference paths, a new approach is first developed based on a proper local coordinate transformation. Then, an alternative solution based on model predictive control is developed to solve the challenge mentioned above in an optimal way. Two linearization methods for the nonlinear model, based on current measurements and steady-state values of time-varying terms, are given. The result shows that both of these two approaches can greatly reduce the tracking error near the intersections of different reference paths. For the trailer far from the reference, a controller focusing on heading control is then proposed. Two methods are then given to eliminate the steady-state error for curved reference path. One is to add a feed-forward part in the control law. The other is to use tractor steering rate as the control input instead of steering angle. The steady-state error may exist when there is a bias in the measurement and process. An additional integrator is added in the model and adaptive integral separation is applied to accumulate position error only when it is close to zero. The proposed approaches are verified by simulations. To reduce the sensor cost, an estimator based on adaptive unscented Kalman filter is proposed when the positions of the trailer are measurable. Simulation results verify the proposed algorithm.