Like many college students, Jackal is still learning to make his own way in life. The little rover is part of a project at the University of Texas' Autonomous Mobile Robotics Laboratory, teaching robots to navigate outdoor terrain using artificial intelligence.
According to its teacher, third-year computer science undergrad Luisa Mao, there have been some bumps along the way.
There was the time Jackal ran off, crashed into a curb and fell in the bushes. Mao said she tried to stop it. But Jackal doesn’t always listen.
Then there was the incident when Jackal rammed into Mao during an experiment to see how well it could find its way around people.
“I think of it as a pet that doesn’t do what it’s trained to do all the time,” Mao said.
Yes, Jackal’s still figuring things out, but so is its teacher.
Mao is one of about 200 students spread across a dozen UT robotics labs. Her lab, AMRL, is funded through industry sponsors — like Amazon, Bosch — and a grant from the U.S. Army Research Laboratory. AMRL specializes in developing AI robots that can go it alone.
More on tech: How robotics research at UT-Austin is working to create real-world impact
As Mao sees it, the robots at AMRL could one day become food delivery droids, first responders or wilderness rescuers. In some cases, autonomous robots are already being used to deliver food, but those bots rely on extremely detailed maps to get around.
“We’re looking at unknown environments, so it’s a more difficult problem to solve,” Mao said. “It’s going to take a long time to develop, years probably, both from the perspective of creating the technology and adapting cities to robots.”
Mao’s experiments revolve around a seemingly simple but practically challenging question: How can Jackal learn to understand its surroundings?
What a robot knows
Sensors help Jackal see the world. But before it can roam parts unknown, the little robot needs to become good at a few basic tasks. The biggest one is distinguishing between different kinds of terrain.
“This thing needs to understand, like, what is grass?” Mao said. “What is sidewalk? Which one should I prefer over the other?”
Because of that, Mao’s research is a little different from other wheeled, autonomous vehicles.
Self-driving cars, for example, use Light Detection and Ranging, or lidar, to measure the distance between obstacles, explained Zhiyun Deng, a first-year doctoral student and one of Mao’s peers in the lab.
But lidar can’t tell a robot what the world looks like except to describe the shape of things.
“For autonomous robots,” Deng said, “we also need a camera to see: What's the texture and colors of the terrain that we need to navigate through?”
Using video, Jackal can see the world around it. Then AI enables the rover to make decisions about where it wants to go.
The bot has brains
In a courtyard near the laboratory on Friday, Jackal hummed. It was hot, and fans inside the rover kept it from overheating.
“One second,” Mao said, typing on her laptop. “I’m loading his autonomous navigation stack.”
Green code flickered across the black screen. These are the brains behind the robot.
Jackal uses a specific kind of AI, Mao explained: Through machine learning, the rover can identify different terrains and decide which ones are best for travel.
Jackal’s brain is made from layers of nodes, also called neurons. Together, they form a neural network stored on a memory drive. In total, the robot’s brain takes up about 2 gigabytes. For reference, a baseline iPhone comes with 64 gigabytes of storage, and AI large language models like ChatGPT require hundreds of gigabytes.
In a few minutes, a new screen appeared on Mao’s laptop.
“So this is what the robot sees,” she said.
A top-down view of the ground in front of Jackal showed the sidewalk in one color and the grass in another. On top of that, the robot charted a series of potential courses.
Mao set a goal for Jackal: get to the end of the courtyard using the sidewalk. She pressed a button, and Deng took his hand off the PlayStation controller they used to get Jackal in place.
“OK,” Deng said. “We switched to autonomous mode.”
Jackal whirred, jerked to a start, then plodded slowly into the distance.
This article originally appeared on Austin American-Statesman: An AI robot is roaming UT Austin — and its name is Jackal.
