Obstacle Filtering Algorithm for Control of an Autonomous Road Vehicle in Public Highway Traffic

This paper presents an obstacle filtering algorithm that mimics human driver-like grouping of objects within a model predictive control scheme for an autonomous road vehicle. In the algorithm, a time to collision criteria is first used as risk assessment indicator to filter the potentially dangerous obstacle object vehicles in the proximity of the autonomously controlled vehicle. Then, the filtered object vehicles with overlapping elliptical collision areas put into groups. A hyper elliptical boundary is regenerated to define an extended collision area for the group. To minimize conservatism, the parameters for the tightest hyper ellipse are determined by solving an optimization problem. By excluding undesired local minimums for the planning problem, the grouping alleviates limitations that arise from the limited prediction horizons used in the model predictive control. The computational details of the proposed algorithm as well as its performance are illustrated using simulations of an autonomously controlled vehicle in public highway traffic scenarios involving multiple other vehicles.