Martin Asenov (cohort student representative)
Modelling spatiotemporal phenomena, like gas concentration, wind speed or temperature, is a challenging task due to the dynamics of the process and the limited number of possible measurements. Active sensing has been the predominant solution, incorporating Gaussian Processes to model the underlying process, and Bayesian Optimisation to efficiently select a new sampling point. Actively steering a drone to maximize the information collected from an onboard sensor is still not commonly used, however, due to different weight, power and computational constraints. In this work, we implement active sensing on a drone with onboard computer and assess its performance. In the first set of experiments we evaluate the performance of existing algorithms with synthetic data using the onboard computer. Next, we demonstrate an autonomous flight of locating multiple flash lights using off-the-shelf light sensor. Finally, we show an integration with a novel broadband spectrometer for detecting and measuring concentration of multiple gases, developed by collaborators at Heriot-Watt University. These results have implications for using drones as a general sensing platform in an intelligent way aiming to maximize the information collected during flight.