Let's meet Disney's new Olaf and find out how he works

March 16, 2026, 4:30 PM · Last week, Walt Disney Imagineering invited me to spend some time with one of Disney's coolest characters.

WDI's new animated Olaf blew up the Internet when it debuted last fall at Disneyland Paris' press event for its upcoming World of Frozen land. My video of the walking, talking snowman animatronic has been seen millions of time across all platforms. But Disney did not allow invited reporters to get too close to its new creation.

Until now.

As you can see, Disney allowed me to get up close with Olaf while talking with Imagineers who helped bring this character to life. First, let's be clear about what Disney has accomplished with Olaf. While the true-to-scale character certainly impresses with its physical size and capabilities, we have not yet arrived in Westworld. Disney's Olaf is not some sentient robot that can walk and talk wherever and however it chooses.

Like other WDI robotic creations, Olaf's movement is controlled by an operator or pre-programmed for a show performance. And Olaf's voice is not synthesized but is instead a collection of recorded vocal performances by Josh Gad, who voices Olaf in Disney's "Frozen" animated films. But Olaf would not appear so life-like and true to his on-screen character if operators had to control manually its every step and facial expression. For that, Disney has developed technology that enables its animatronics to "learn" how to handle that immensely important small stuff on their own.

Disney has been sculpting and programing life-like animatronics for decades. But how can Disney now communicate such fine detail and robust movement at the same time, with mechanisms contained within a character that stands just a couple of feet tall?

The simplistic answer is, of course, "Disney magic." Arthur C. Clarke once said that "any sufficiently advanced technology is indistinguishable from magic." Yet what seems like magic to a guest is in reality a collection of tech challenges for Imagineers to address and solve.

If you want to dive into the mechanics of Olaf, including Disney's use of the NVIDIA Newton Physics Engine, spend some time with the study linked in my previous post, Disney shares a look inside how it made its new Olaf. Or, let's hear from the Imagineers I spoke with last week.

"The real breakthrough for us and our ability to deliver these characters in true-to-form, true-to-screen experiences, has always been this branch of artificial intelligence called reinforcement learning," Kyle Laughlin, SVP, R&D and Technology and Engineering, Walt Disney Imagineering, said. "It takes all of that time that would have been necessary to hand-program and compresses that into days and weeks."

"When Olaf first started walking, his feet made a really loud noise on the ground," Josh Gorin, Executive, R&D and Disney Live Entertainment Innovation, Walt Disney Imagineering, said. "They were kind of loud. So we could go back into the reinforcement learning model and actually give rewards for walking quieter, and he could learn how to step more quietly. We were able actually to reduce the sounds of his footfalls through reinforcement learning about changing any of the hardware."

I had to ask how, exactly, Imagineers "reward" a robot when learning, because it's not like they are feeding Olaf treats.

"What we literally do is you take a full 3-D physics simulation of the robot, of the motors, the wires, the bolts, the screws, you bring it into an environment that is full physics simulation, so it has gravity," Gorin said. "We can bring in different surfaces for him to walk on, and then over time, you basically reward the model. Try to stand up, try to walk, try to walk like Olaf. And that's where this animator training data is so important, where the animators who actually worked on Olaf, who worked on the film, give training data for how to move like Olaf, and then over time, over tens of millions of iterations, it learns how to do what we want it to do, which is move and express and emote like Olaf. So it is the computational equivalent of giving treats. But in this case, it can happen tens of millions of times over the course of days, as opposed to years and years and years."

"The basics of just walking used to take us so long that now that time that used to be dedicated to that activity, we can focus on the actual artist's intent and the details that make him so life-like," Laughlin said. "Part of what I think is really magical about Olaf, obviously, is he's always looking to try and make eye contact with you, and... that is incredibly taxing on this actuator that's in his neck.

"It's got, obviously, such a big head and we have a small costume, so we use the thermal reinforcement learning to be able to understand the temperature in his neck, because so much of the of the of the magic is looking up at you that the motor overheats very quickly, and shuts the robot down. But the team put some guardrails in place through reinforcement learning to be able to change the way that Olaf might be able to interact with you in real time - taking that information from that actuator and beginning to cool it down on its own, as if he had his own ice pack inside."

Because no one wants to see Olaf melt.

Olaf appears inside the World of Frozen land at Disney Adventure World in Paris (the soon-to-be-former Walt Disney Studios Park) and Hong Kong Disneyland. But now that Disney cracked the code for making engaging characters of this size, he soon might be joined by other next-generation friends in Disney's theme parks around the world.

"You can expect to see a whole suite of characters from our biggest franchises that all can interact with one another in the future," Laughlin said. "So nothing to talk about yet, but we're excited about the potential for a whole suite of new characters coming out."

For our weekly update of Disney and theme park news, please sign up for Theme Park Insider's weekend newsletter.

Replies (0)