Path planning algorithms for atmospheric science applications of autonomous aircraft systems

Abstract

Among current techniques, used to assist the modelling of atmospheric processes, is an approach involving the balloon or aircraft launching of radiosondes, which travel along uncontrolled trajectories dependent on wind speed. Radiosondes are launched daily from numerous worldwide locations and the data collected is integral to numerical weather prediction.

This thesis proposes an unmanned air system for atmospheric research, consisting of multiple, balloon-launched, autonomous gliders. The trajectories of the gliders are optimised for the uniform sampling of a volume of airspace and the efficient mapping of a particular physical or chemical measure. To accomplish this we have developed a series of algorithms for path planning, driven by the dual objectives of uncertainty and
information gain.

Algorithms for centralised, discrete path planning, a centralised, continuous planner and finally a decentralised, real-time, asynchronous planner are presented. The continuous heuristics search a look-up table of plausible manoeuvres generated by way of an offline flight dynamics model, ensuring that the optimised trajectories are flyable. Further to this, a greedy heuristic for path growth is introduced alongside a control for search coarseness, establishing a sliding control for the level of allowed global exploration, local exploitation and computational complexity. The algorithm is also integrated with a flight dynamics model, and communications and flight systems hardware, enabling software and hardware-in-the-loop simulations. The algorithm outperforms random search in two and three dimensions. We also assess the applicability of the unmanned air system in ‘real’ environments, accounting for the presence of complicated flow fields and boundaries. A case study based on the island South Georgia is presented and indicates good algorithm performance in strong, variable winds. We also examine the impact of co-operation within this multi-agent system of decentralised, unmanned gliders, investigating the threshold for communication range, which allows for optimal search whilst reducing both the cost of individual communication devices and the computational resources associated with the processing of data received by each aircraft. Reductions in communication radius are found to have a significant, negative impact upon the resulting efficiency of the system. To somewhat recover these losses, we utilise a sorting algorithm, determining information priority between any two aircraft in range. Furthermore, negotiation between aircraft is introduced, allowing aircraft to resolve any possible conflicts between selected paths, which helps to counteractany latency in the search heuristic.

More information

Identifiers

ORCID for Andras Sobester: orcid.org/0000-0002-8997-4375

Catalogue record

Export record