One of the more unexpected moments at NVIDIA GTC 2026 had nothing to do with chips or large language models. A robotic Olaf — the beloved snowman from Disney's Frozen — walked onto the keynote stage alongside CEO Jensen Huang, capping a multi-year collaboration between Walt Disney Imagineering, NVIDIA, and Google DeepMind.

Built to Move Like a Snowman

The robot is the result of deep reinforcement learning running on NVIDIA GPUs through Disney Research's Kamino simulator, a GPU-accelerated physics solver that runs thousands of parallel training environments simultaneously on a single GPU. Rather than manually programming Olaf's gait, animators provided training data to teach him his signature shuffle.

The challenge was deliberately hard: Olaf had to learn to walk not just on flat ground, but on the deck of a boat. Through simulation, he achieved that in hours — the same task would take years for a human child.

The project builds on the Newton Physics Engine, an open-source simulation framework jointly developed by Disney Research, NVIDIA, and Google DeepMind.

Park Debut: March 29

The GTC appearance wasn't just a demo. Olaf is set to make his guest-facing debut on March 29 at World of Frozen, Disneyland Paris, as part of the "Celebration in Arendelle" show — performing on a boat in the park's lagoon.

Disney Imagineering SVP Kyle Laughlin confirmed that Kamino is being evaluated for training future robotic characters. The Olaf robot joins a lineage of free-roaming droids already operating at Star Wars: Galaxy's Edge, but marks the first character trained entirely through GPU-accelerated simulation for real-world deployment.