Artificial intelligence lets glider find its own thermals
— Sign up for the new EW Daily Newsletter, for latest news and products —
Hampshire-based lab Roke Manor Research is developing an artificial intelligence (AI) system that looks at clouds to allow autonomous gliders to find lift in the sky.
“The primary objective is to harvest atmospheric energy to prolong mission duration,” said Roke. “This is achieved by automated on-board energy aware planning, whereby areas of rising air are identified and exploited, and regions of sinking air are avoided.”
Models are used to predict vertical air movement from two sources: the wind blowing over physical land features, and the sun warming land surfaces causing ‘thermals’.
“At the core of the software lies a map-based ‘blackboard’ knowledge management architecture,” said Roke. “This is an agent-based architecture, featuring decoupled knowledge source and data processing agents.”
The central blackboard is the synthetic equivalent to a World War II air raid planning table.
Some agents post predicted vertical air velocity data onto its surface, others look at the data to create a flight plan.
In trials, Roke used 3D atmospheric wind velocity data from the Met Office, adding in the effect of local terrain on the lower reaches of that data set using computational fluid dynamics.
Solar-induced vertical air movement was predicted by automating a technique used by human glider pilots.
Thermals form over land which is warmer than surrounding land, so pilots look for large areas likely to heat more quickly than adjacent areas – a substantial car park next to a lake for example.
Once formed, a cloud of a characteristic shape develops at the top of the thermal.
Different surface types – for example: grassland, woodland, tarmac and urban – were observed for two years to develop an ‘energy balance model’ which predicts how land heats the air above it.
According to Roke, on average summer’s day tarmac is 8°C hotter than grass.
This knowledge was applied to surface type data from the trial area to identify likely stating points for thermals.
To identify actual thermals above candidate surfaces “a novel ‘cloudscaping’ video image processing algorithm, detects thermal columns of rising air from a live video feed, providing an up-to-date picture of local conditions,” said Roke.
Using data presented on the blackboard, further agents find a route to the required destination that loses least height, based on information provided by mission requirement agents.
“The ‘planners’ reason over the blackboard world model to produce detailed waypoint sequences for the flight management system to use,” said Roke.
Further work may include having additional agents that can communicate with similar nearby aircraft to share data.
Potential applications include extending the flight range of powered military, communication, and surveying un-manned aerial vehicles, said Roke.Tags: artificial intelligence, developing, system