A paper published in this week’s Nature describes how scientists from universities in the US and Italy trained an algorithm to control a glider to navigate thermals.
Birds don’t always go from A to B by flapping their wings. Sometimes they take advantage of rising currents of warm air known as thermals. How they do this isn’t particularly well known but scientists are using AI to learn every avian trick and then apply that knowledge to teach aircraft to do the same.
This isn’t the first time this has been done, but it is the first time data from real flights has been used to improve the performance of AI in the field.
The scientists trained their algorithm in flight simulators and in more than 200 flights over California. They learned that factors such as vertical wind acceleration and side-to-side torque were important to help gliders fly smoothly and this may be the same for birds too.
The gliders were trained using reinforcement learning, a type of machine learning where a software agent learns how to behave by performing actions and seeing the results.
“This paper is an important step toward artificial intelligence—how to autonomously soar in constantly shifting thermals like a bird,” said Terry Sejnowski, a member of the research team from the Salk Institute for Biological Studies.
“I was surprised that relatively little learning was needed to achieve expert performance.”
The work may provide a way in future years for autonomous vehicles to take advantage of thermals rather than rely on powered flight.
For all the details go to the Nature website to read the paper entitled ‘Glider soaring via reinforcement learning in the field’.