Realtime Determination and Prediction of Aircraft Trajectories Using Limited Sensor Data

Dragonfly Pictures, Inc. (DPI) and UPENN propose to develop a novel real-time trajectory prediction tool called Trajectory Tree Prediction Planner (TTPP). TTPP is based on two ideas. First, since it is impossible to robustly predict a single trajectory of an aircraft if its full state and intent is unknown, we advocate generating a set (tree) of trajectories. Each of the trajectories will be associated with the probability of occurrence, and the cumulative probability of predicted trajectories will be near one. Second, while the prediction of multiple trajectories will be even more computationally expensive than the prediction of a single trajectory, we will generate these trajectories using our previously developed incremental planning algorithms (e.g., D* Lite, Anytime D*, Adaptive A*). These algorithms are specifically designed for solving repeated planning problems in real-time and compute the new paths orders of magnitude faster by re-using their previous planning efforts. These algorithms have been shown to work well on various unmanned ground and air vehicles and will therefore be well-suited to the problem of generating the tree of future trajectories in real-time.