Particle filtering for online motion planning with task specifications
Karl Berntorp, Stefano Di Cairano · 2016
A probabilistic framework for online motion planning of vehicles in dynamic environments is proposed. We develop a sampling-based motion planner that incorporates prediction of obstacle motion. A key feature is the introduction of task specifications as artificial measurements. This allows us to cast the exploration phase in the planner as a nonlinear, possibly multimodal, estimation problem, which is effectively solved using particle filtering. For certain parameter choices, the approach is equivalent to solving a nonlinear estimation problem using particle filtering. The proposed approach is illustrated on a simulated autonomous-driving example. The results indicate that our method is computationally efficient, consistent with the task specifications, and computes dynamically feasible trajectories.