Autonomous Plume Tracking Unmanned Aerial Vehicles

The project aims to address the question of how an unmanned aerial vehicle can used to track the sources of air pollution. The project combines the use commercially available low cost air quality sensors, and utilizes machining learning approaches to develop planning and source seeking strategies. Within this question the project will examine how a machine learning approach accommodate the particular sensor technologies with their known/estimated levels of uncertainty.

The approach will utilize laboratory testing to calibrate and benchmark a range of commercially available low cost sensor hardware against industry/academia standard high fidelity sensors. The approach will also examine the influence of the vehicle itself (e.g. occlusion of sensors, downwash effects) on the data captured. In parallel with the physical engineering approaches, machine learning approaches to plume tracking will be reviewed with an eventual ambition of implementing a smart approach to plume tracking using onboard sensor data and trained path planning strategies. Within the machine learning phase different levels of modeling fidelity will be examined, from traditional Gaussian plume type models, to more advanced in-house lattice-boltzman codes that can incorporate the influence of obstacles (landscape, buildings) on flow and plume structure.

The novelty in the content lies in combining recently available low-cost sensors that are potentially flight-worthy, with an advanced machine learning approach.