Model-predictive active steering and obstacle avoidance for autonomous ground vehicles

This paper presents a model-predictive approach for trajectory generation of unmanned ground vehicles (UGVs) combined with a tire model. An optimal tracking problem while avoiding collision with obstacles is formulated in terms of cost minimization under constraints. Information on obstacles is incorporated online in the nonlinear model-predictive framework as they are sensed within a limited sensing range. The overall problem is solved online with nonlinear programming. For the local path regeneration upon detecting new obstacles, the cost function is augmented using the obstacle information in two methods. The first method uses the distance from the UGV to the nearest detected obstacle, and the second method uses the parallax information from the vehicle about the detected obstacles. Simulation results in cluttered and dynamic environments show that the modified parallax method effectively reflects the threat of the obstacles to the UGV considering the dimension and state variables of the vehicle, showing clear improvements over the distance-based methods.