Online plan modification in uncertain resource-constrained environments

This paper presents an approach to planning under uncertainty in resource-constrained environments. We describe our novel method for online plan modification and execution monitoring, which augments an existing plan with pre-computed plan fragments in response to observed resource availability. Our plan merging algorithm uses causal structure to interleave actions, creating solutions online using observations of the true state without introducing significant computational cost. Our system monitors resource availability, reasoning about the probability of successfully completing the goals. We show that when the probability of completing a plan decreases, by removing low-priority goals our system reduces the risk of plan failure, increasing mission success rate. Conversely, when resource availability allows, by including additional goals our system increases reward without adversely affecting success rate. We evaluate our approach using the example domain of long-range autonomous underwater vehicle (AUV) missions, in which a vehicle spends months at sea with little or no opportunity for intervention. We compare the performance to a state-of-the-art oversubscription planner. Planning within such domains is challenging because significant resource usage uncertainty means it is computationally infeasible to calculate the optimal strategy in advance. We also evaluate the applicability of our plan merging algorithm to existing IPC domains, presenting a discussion of the domain characteristics which favour the use of our approach