Postdoc working on AI for astronauts

Ulubilge Ulusoy is advancing the science of artificial intelligence to help astronauts on future missions to Mars. 

Although such a flight is years away, AI systems will be critical to assisting astronauts as they journey beyond Earth’s orbit. 

“These are problems we need to understand now,” Ulusoy said. “The International Space Station works because we have a huge group of humans in ground control helping these systems to function. When you go to Mars and the time delays for communications can be up to 24 minutes one way, it’s impossible for NASA to check things in real time. You need AI-based systems ready to keep the crew alive and the habitat functional.”

Ulusoy is designing methods to test human responses to AI systems within a framework where AI systems assist humans and humans also assist AI. The goal is to better understand how people can best interact with them.

He is the inaugural Smead Distinguished Postdoctoral Associate at the University of Colorado Boulder. The one-year position provides funding for leading PhD graduates conducting aerospace research.

Ulusoy uses sensors to collect an array of physiological data as humans interact with AI systems in real-time, including heart rate variation, perspiration, breathing rate, and cognitive load.

Such objective data can reveal how trustworthy and useful human users find AI systems as well as the best methods AI can use to provide assistance.

“These systems are getting better, but we often ignore them anyway. You sometimes have Siri or Google pop up to say something and you’re like, ‘Not now, Siri.’ These new AI systems have machine learning technology so when you provide counter feedback, it can learn from that,” he said.

Ulusoy is conducting pilot tests this semester with the hope of beginning human trials in the spring. He and fellow Postdoctoral Associate Patrick Pischulti are working with Katya Arquilla, an assistant professor in the Ann and H.J. Smead Department of Aerospace Engineering Sciences, and an expert on how humans interact with and adapt to complex systems.

The research is a long-time goal of Ulusoy. He first began working on AI/astronaut interactions in 2020 during his PhD program. At that time, its potential was largely theoretical.

“I wanted to present a framework on paper, but in my third year, commercial, off-the-shelf large language models came out. It is a really exciting time to move forward,” he said.

The experiments this spring will see human test subjects working with a robotic arm to complete a physical task. Ulusoy will monitor the human test subject responses to see how well they work together.

“Being able to do human factors work, specifically collective physiological data is really helpful,” he said. “Implementing real-time, objective measurements will bring this to the next level.”