Abstract:

Autonomous driving in urban and highway scenarios involves a set of predefined requirements that the vehicle should obey, such as lane following, safety distances to surrounding vehicles, and speed preferences. We have previously shown that by interpreting the motion-planning problem as a nonlinear non-Gaussian estimation problem, we can leverage particle filtering to determine suitable vehicle trajectories. In this paper, we validate our proposed motion planner using scaled vehicles. We show that our motion planner is capable of determining safe and drivable trajectories for a number of challenging scenarios, and that the trajectories can be accurately tracked by a lower-level nonlinear model predictive control scheme.