This article was written by Amii Fellow and Canada CIFAR AI Chair Marlos C. Machado, who serves as an Assistant Professor of Computing Science at the University of Alberta. He designs algorithms that learn abstractions for better credit assignment, generalization, and exploration in reinforcement learning.
In November 2023, Machado delivered a fascinating presentation on “Flying Balloons in the Stratosphere of AI” at the Jasper Dark Sky Festival – an annual festival celebrating solar wonder and stellar delights, in the heart of one of the most breathtaking dark sky preserves, Jasper National Park.
Below, Machado reflects on his experience at the festival, connects stratospheric balloons to space science, and reinforces the interdisciplinary nature of science.
In this talk, I delved into our work developing an artificial intelligence (AI) algorithm to serve as a high-performing flight controller for superpressure balloons.
Our solution surpassed human-designed control methods, demonstrating its effectiveness in tackling real-world autonomous control problems where conventional approaches and human intervention fall short. Our solution was not only more accurate but also more power efficient.
For the Jasper Dark Sky Festival, I presented an overview of the fundamental AI concepts employed in the project, providing insights into the development process and shedding light on the intriguing behaviours learned by the AI system.
An example I love to share is that, simply by “experiencing” physics in the world, our system learned to store potential energy to become more power efficient. It did so by controlling the altitude it was in at the end of the day when solar power was not available.
Superpressure Balloons and Space Science
This talk was particularly fitting for the Jasper Dark Sky Festival because superpressure balloons are closely tied to space science:
The stratosphere is a unique region, positioned between the altitudes at which satellites and aircraft typically operate.
It is characterized by its proximity to the ozone layer and shares some similarities with the space environment.
It is within this realm that superpressure and high-altitude balloons reside, offering diverse applications such as weather monitoring, aerial photography, telecommunications, and technology validation for long-duration space missions.
As the Canadian Space Agency emphasizes, “Stratospheric balloons offer a cost-effective platform for advancing space science”.
Eyes on the Skies
Presenting our work in using AI to control superpressure balloons at the Jasper Dark Sky festival was particularly thrilling, given the project’s broad appeal. As David Bourgeois wrote in 1784, “The idea of taking to the air, of flying through the sky and navigating through the ether, has always appealed so strongly to mankind that it has appeared in numerous classical legends and folktales from the remotest antiquity”; effectively navigating balloons was a historic challenge.
This is a topic that underscores the interdisciplinary nature of science, emphasizing, as I did in the talk, how science is a multidisciplinary endeavour, and how AI can be used to empower us to greater heights (I couldn’t resist the pun).
As Richard Holmes wrote, “Perhaps, we need the three things that a scientific culture can sustain: the sense of individual wonder, the power of hope, and the vivid but questing belief in a future for the globe.”
Watch Marlo’s talk about his work applying reinforcement learning to autonomously navigate stratospheric balloons in this previously recorded AI Meetup.