Path Planner for Dynamic Environments

Despite excellent recent progress in path planning, current state-of-the-art planners are still incapable of operating robustly in environments that are both cluttered and highly dynamic. The challenge is two-fold. First, predicting the trajectories of dynamic obstacles (i.e., cars, humans, aircraft) is very noisy. As a result, to achieve robust execution, perception modules need to provide a richer representation of predictions while planning modules need to account for the uncertainty in the predictions. Second, planning with dynamic obstacles is computationally complex because it requires adding time as an additional dimension while paths in dynamic environments need to be constantly re-computed. In Phase I, we have developed a novel path planner that addresses these challenges. It includes a novel representation of dynamic obstacles that can be used to generate safe paths efficiently and is general enough for use with different sensor suites. The planner also uses a novel data structure - time-bounded lattice. The lattice can be used efficiently with our previously developed re-planning algorithms suitable for constant re-planning. In Phase II, we propose to extend the planner to support planning for UAVs, transition it into the DPI real-time control software and demonstrate its performance on real test flights of DPI helicopters.