The new Research & Development episode of We Call It Imagineering showcases how Imagineers use artificial intelligence (AI) to create advanced robots, bringing various characters to life.
Disney Robots Using AI to Learn
Imagineers demonstrated the technology using a walking robot that they have previously shared videos of. Imagineers use reinforcement learning to train robots. Reinforcement learning is a branch of AI that allows them to take animated figures, bring them into a simulation, and then teach them how to walk just like humans learn.
The robots can complete 10,000 hours worth of training in mere days thanks to the simulation. They imitate artist-provided motions several times, slowly becoming better and achieving more polished movements. They can learn multiple motions that an animator asks them to do.
Of course, getting a robot to learn a movement in simulation is different than getting the hardware to actually recreate it in the physical world. Imagineers had to take time to understand and tune actuators so the robot wouldn’t break if it fell down.
The walking robot was the precursor to the Star Wars BDX droids that have now appeared worldwide at Disney Parks. These droids took less than a year to go from concept to appearances at Disney Parks. They use spatial AI facilitated by cameras to “see” and understand their environment, including the positions of guests around them.
Two cameras in each droid are their “eyes and ears,” helping them become more autonomous. Disney partnered with AI company NVIDIA, who builds revolutionary computer chips for the robots, allowing Imagineers to focus more on the connections they make with guests. NVIDIA also created a simulation framework called Newton and the Camino simulator.
We Call It Imagineering showed Imagineers practicing and recording a show sequence. The droids are initially controlled by Imagineers using remotes, but after the show is practiced and recorded, with the droids learning the movements, they are able to run the show back without remote controls.
Another robot showcased in the episode is a ballerina that can pirouette. It has motors at its joints, although Imagineers pointed out that’s not always possible for animatronics.
The end product of reinforcement learning is a neural network in the robot which can evaluate the robot’s current state and decide its next action. For example, if it knows it’s about to fall over, it can take a step to correct itself and remain standing.
Behind the scenes, Imagineering has developed their own AI tool inspired by Iron Man’s JARVIS. This is an AI ecosystem that gives Imagineers access to their collective intelligence. They have digitized 70 years of data and made it available at Imagineers’ fingertips.
Imagineers emphasized the “technology behind the magic” and the hope of creating a seamless experience for guests that connects movies, streaming, gaming, and Disney Parks.
Watch the Research & Development episode of We Call It Imagineering below.
For the latest Disney Parks news and info, follow WDW News Today on Twitter, Facebook, and Instagram.