An adaptive sampling method for generating boundary scenarios of UAV swarms based on Gaussian process regression

In the field of autonomous system testing, the dynamic and interactive nature of multi-agent systems, such as UAV swarms, poses substantial challenges for effective testing and validation. Traditional boundary scenario generation techniques struggle to comprehensively identify boundary scenarios in UAV swarm testing, primarily due to the sparsity and complexity inherent in high-dimensional scenario spaces. To overcome these limitations, this paper proposes an adaptive sampling process that is guided by a Gaussian Process Regression (GPR) surrogate model. This method leverages predictive insights from the surrogate model to dynamically adjust the balance between exploring and exploiting the scenario space, aiming to identify critical transition areas in UAV swarm performance. By focusing on distinguishing boundary scenarios that demarcate distinct performance modes, this approach optimises testing efficiency and deepens the understanding of UAV swarm behavior under varied operational conditions. The efficacy of this novel method in testing complex autonomous unmanned systems is validated through a series of simulation experiments involving UAV swarm search tasks.