Risk Aware SUAS Path Planning in an Unstructured Wildfire Environment

This paper addresses path planning for small unmanned aerial vehicles (SUAS) in an unstructured environment with poorly understood and time varying obstacles. These environmental characteristics manifest in an evolving wildfire, which is used as the test-case for this work. An evidential framework is employed to estimate the current state of wildfire and the resulting heat aura at flight level. This approach accounts for ignorance in estimation that can result from frequent conflict among sensors operating in a harsh environment combined with a computational forecasting agent that must operate with poor models of fire evolution. The objective of SUAS mission design is to visit such points of high conflict in order to provide additional situational awareness. A novel unsupervised classification algorithm based on statistical formalism is developed to identify distinct keep-out zones within the estimated heat aura. Boundaries of identified obstacles are modeled as probabilistic barriers. The planning problem is posed as a chance-constrained optimal control problem, which is transcribed to a nonlinear program via pseudospectral discretization. It is shown that by actively varying the probability of violating obstacle boundaries, a family of solutions can be generated that elicit the risk associated with each mission design.