A POMDP for Optimal Motion Planning with Uncertain Dynamics

This paper describes an approach to the problem of optimal motion planning with uncertain dynamics. This problem can occur whenever a vehicle suffers damage or when the environment makes the effect of motion actions unpredictable and potentially risky. We address in particular the case of aircraft with structural and/or control surface damage. The goal in this problem is to find an optimal plan for emergency landing, which might entail additional exploration of the aircraft flight envelope. However, this exploration is risky, and must be balanced by possible improvements in the resulting emergency landing plan. Evaluating the risk and potential benefit of exploration allows focusing on the fraction of the envelope that is beneficial, given the current situation (obstacles and possible landing sites). This reduces both the risk and the cost of exploration. The paper surveys previous work on optimal motion planning and flight envelope exploration. It shows how the problem of interleaving both tasks can be modeled as a Partially Observable Markov Decision Process where local dependencies between control states are modeled using Markov Random Fields.