Just like an eagle, this autonomous glider can fly on thermal currents
An eagle soaring may look majestic but in technical terms, there is some impressive physics happening “under the hood” when they do. Specifically, eagles and other soaring birds take advantage of the upward currents of warm air, known as thermals, to help them more easily sail through the sky. What scientists don’t know, however, is how these birds discover and navigate said thermals. It turns out that artificial intelligence can help — and it could offer an assist to drones as an added bonus.
“This is a big challenge, as it is very difficult to conduct controlled experiments with soaring birds,” Jerome Wong-Ng and Gautam Reddy, two researchers from the University of California, San Diego, wrote in an email to Digital Trends. “Our approach was to instead teach a learning agent to soar in a realistic environment and see if this tells us something about how birds soar.”
This teaching was carried out using a type of machine learning called reinforcement learning. This type of A.I. creates A.I. agents which learn behavior based on the results of trial and error experiments. In this case, the researchers kitted out a glider with a flight controller able to implement the reinforcement learning-based instructions. Soaring to heights of almost 2,300 feet, the glider was able to figure out how to navigate atmospheric thermals autonomously.
“On a technical level, reinforcement learning hasn’t been applied to train agents to learn in the field,” the researchers continued. “In the field, the number of training samples we have is really low, and we have to come up with ways of using all available training data. There were also technical advancements regarding how to measure the local wind environment near the glider using onboard devices.”
In terms of practical applications, the researchers think their new navigational strategy could be employed to develop unmanned aerial vehicles (UAVs) able to fly for long periods of time without needing to recharge. In addition, it might be useful for creating an autopilot-style “recommendation system” for novice glider pilots.
“In this work, we focused on how to find and navigate a single thermal,” Wong-Ng and Reddy said. “But migrating birds glide from one thermal to another, and how to do this efficiently is a line of work we plan to explore in the future. Another line of research is to track soaring birds and figure out if their navigational strategy is similar to the one we’ve found in our study.”
Along with the University of California, San Diego, other educational institutions involved in this research included the Salk Institute for Biological Studies and the Abdus Salam International Center for Theoretical Physics in Trieste, Italy.
A paper describing the research was recently published in the journal Nature.
- A.I. players master ‘Quake III Arena,’ manage to outperform humans
- By studying patient data, A.I. can limit toxicity in cancer treatment
- How 3,000 streetlights turned San Diego into America’s smartest city
- IBM’s Holodeck-style classroom tech makes language-learning apps look primitive
- Machine learning? Neural networks? Here’s your guide to the many flavors of A.I.