Data-driven risk assessment and multicriteria optimization of UAV operations

This paper introduces a novel data-driven risk metric and assessment method for UAVs operating in environments typically encountered in civilian applications. A truly “data-driven risk measure” is derived through a probabilistic formulation that not only accounts for the intrinsically stochastic nature of the considered environmental factors (such as weather and signal strength), but also incorporates extrinsic prediction uncertainties originating from the geographical sparsity of data collection sources. We present a data-driven modeling of the stochastic environmental factors using Gaussian process-based function approximations. Notably, the proposed mathematical definition of the risk metric is based on the probabilistic predictions of such a Gaussian process model and introduced through a path-integral formulation. The problem of minimizing operational risk for multiple UAVs in partially unknown environments is then defined in a multicriteria optimization framework to address the trade-off between the path-integral risk measure and classical path-efficiency (distance). We show that such approach can be embedded into current standard risk assessment methods which could be easily integrated into UAVs traffic management initiatives. We analyze the results through a number of simulations, including realistic scenarios.