Gaussian-based kernel for multi-agent aerial chemical-plume mapping

This paper presents a multi-vehicle chemical-plume mapping process that incorporates onboard wind speed and direction estimation. A Gaussian plume model is exploited to develop the kernel for extrapolating the measured data. Compared to the uni-or bi-variate kernels, the proposed kernel uses the estimated wind information to refine the chemical concentration prediction downwind of the source. This new approach, compared to previous mapping methods, relies on fewer parameters and provides 30% reduction in the mapping mean-squared error. Simulation and experimental results are presented to validate the approach. Specifically, outdoor flight tests show three aerial robots with chemical sensing capabilities mapping a real propane gas leak to demonstrate feasibility of the approach.