Jennifer the Humanoid Robot Learns How to Ski
Aside from being cute, Jennifer the robot is helping advance the field of artificial intelligence more than a chess-playing robot ever could.
The University of Manitoba AAL team at last year's Fira HoroCup. Image: AAL
Of course the humanoid robot that can both ski and play hockey was built by Canadians. Researchers at the University of Manitoba's Autonomous Agents Laboratory have developed a sports-playing robot with a range of motions, including the ability to ice skate and ski both cross-country and downhill.
Dubbed Jennifer, the machine is a continuation of sports-playing humanoids that the team has spent years developing and entering into competitions like the FIRA HuroCup, a kind of robot Olympics where the Manitoba team took first place two years ago
The competition challenges robots to perform a range of tasks, meaning the teams can't simply program the robot to do one thing really well, it has to figure out how to move and interact with different environments as it encounters them.
"It was originally developed for dealing with the breadth of humanoid motion. One of the big things our lab is interested in is doing things like gaits for walking, skating, things like that where you have to deal with the fact that the ground underneath you might be changing," explained John Anderson, the director of the autonomous agents lab.
But it's not just for the fun of seeing a human-shaped robot shimmy through the snow. Anderson, and co-director Jacky Baltes, say the technology that's developed through sports-playing robots could one day lead to mechanized firefighters or rescue bots.
A humanoid shape isn't necessary for a robot designed to complete a single task, like delivering office mail, Baltes told me. It's too expensive and fragile, while a sturdy cart with wheels is much more effective.
"Where the humanoid robot shines is that it works in an environment created for humans and it can do many different tasks," Baltes said.
Sports are actually the perfect way to test out the breadth of motion in humanoid robots for a number of reasons, Baltes said. For one, there's an element of competition that encourages researchers to keep pushing themselves. But it's also a way of really challenging the limits of what robots can do.
"Sports, from an artificial intelligence perspective, require a tremendous amount of intelligence, much more so than people give them credit for," Baltes said.
When it comes to AI, developers long believed that challenging robots with complicated abstract thought tasks were the best way to advance the technology. But while crushing a game of chess or Jeopardy! proved pretty simple for a robot, abilities that we take for granted—like navigating a 3D space or recognizing a loved one's face—are much more difficult.
"We focus on these tasks that most people think are trivial but are very, very hard to do on a computer and require a tremendous amount of intelligence. We just don't notice it because we can do it subconsciously," Baltes said.
In May, Jennifer will get to put her intelligence to the test in another competition, as the university's official entry in the humanoid challenge at the International Conference on Robotics and Automation. And maybe someday in the future, she won't just be skiing down the slopes—she might be rescuing your injured butt off the mountain.