Spatiotemporally Consistent Smooth Speed Profiles for Autonomous Driving

Abstract

Graph-based trajectory planning algorithms such as A*, Dijkstra or RRT use motion primitives to generate obstacle-free trajectories. A common method is the use of piecewise constant acceleration profiles along these primitives to avoid high computational complexity. However, the resulting trajectories can be uncomfortable to drive for the passengers due to the abrupt changes in the acceleration. In this paper, we present three different methods to smoothen given trajectories which are discontinuous in their acceleration profile, while complying with the given lengths as well as times. Due to the latter, safety is still guaranteed while the driving comfort is increased. Our approach allows to decrease the complexity of the trajectory-generation problem by relaxing the driving-comfort requirements on the initial creation and shifting it to a second smoothing step.