So-called long heavy veh icle combinations (LHVCs) are trucks up to 32 meters in length and 80 tonnes heavy. They have the potential to reduce greenhouse gas emissions, traffic congestions and transportation costs for road freight transport. LHVCs are widely used in Canada and Australia today, and their abundance on European roads is expected to increase in the near future. However, an undesired effect of the added towed units is the increase in difficulty to maneuver such trucks on roads and in busy traffic. The increasing complexity for truck drivers to handle trivial tasks, like changing lane, call for advanced assistance functions. The development of advanced driver assistance systems or potentially autonomous functioning trucks can improve traffic safety. This allows a further increase in use of long combination trucks. One crucial component of such a driver assistance system is the ability to plan a safe and smooth trajectory in real-time. In this thesis, a nonlinear receding horizon trajectory generator is proposed for highway driving of an A-double combination type of LHVC. An optimal control problem (OCP) is formulated to define the open-loop constrained optimal trajectories. Optimality is defined as a trade-off between three main components. Firstly, the jerk levels perceived by the driver is to be minimized. Secondly, the lane center ought to be tracked and the velocity of the traffic flow should be followed. And finally, depending on the detected scenario, distance is maintained from a set of fellow road users. Hard constraints are imposed for the actuator limitations, to prohibit the truck to leave the lane boundaries and to limit lateral acceleration levels. Actuation signals are generated for the low-level steering control and a longitudinal velocity tracker. The prediction of the vehicle states is performed using a nonlinear single-track model of the A-double combination with the linear tire slip assumption. The prediction horizon of the optimal trajectory is defined in the traveled distance along the lane center. This allows a natural definition of the road curvature, progressing linear in the prediction. A reformulation of the vehicle prediction model is executed for its incorporation in the OCP. Surrounding vehicles are assumed to drive constant velocity, a spatial prediction model is defined for the trajectories of other road users. A direct multiple-shooting solution technique with a piecewise constant control parameterization is used to obtain a nonlinear program (NLP). Using the ACADO Toolkit, the so-called Real-Time Iteration (RTI) scheme is implemented exploiting a constrained Gauss-Newton solution algorithm. Instead of solving the entire NLP each control interval, a sequential quadratic programming (SQP) technique is synchronized with the sampling of the controller. Each intermediate solution iterate of the SQP is used as a control reference to the vehicle. The solution strategy is separated into two distinct phases, a preparation step and a feedback step. Only the feedback step requires knowledge of the most recent state, a short duration of this step assures minimal feedback delay. The solution strategy to the NLP is implemented in efficient stand-alone C/C++ code, interfaced with Simulink and the motion simulator of the Swedish National Road and Transport Research Institute (VTI) in Göteborg, Sweden. All routines are executed based on a clearly defined set of measurements on the vehicle, the road and the surrounding traffic. The overall control algorithm is tested in closed-loop simulations on two A-double combination models. The vehicle prediction model (also employed in the NLP) and a high-fidelity vehicle model, provided and validated by Volvo Group Truck Technologies. Results of simulations are presented for lane changes, merging actions and evasive maneuvers on low-curvature highways. The trajectory generator successfully controls both plant models for the intended scenarios. Slight model-mismatch is observed between the prediction model and the high-fidelity plant, which has a limited deteriorating effect on the control performance. Execution times of the NLP solution strategy show that the trajectory generator implementation is real-time feasible. In general we conclude that this work successfully demonstrates the applicability of the RTI-algorithm to control the A-double combination for highway maneuvering. (Author/publisher)
Samenvatting