How it differs from traditional animatronics
According to The Verge, Disney park animatronics have so far followed predefined routines or been animated by a hidden puppeteer. Walt Disney Imagineering R&D has presented three new Audio-Animatronics prototypes that instead use the cloud and Amazon Web Services' artificial intelligence capabilities to respond uniquely to guest interactions in real time.
The three characters are named Vilo Vele, Kado Bre'a and Jibu Bre'a and are derived from the Anzellan aliens appearing in the film The Mandalorian and Grogu, released this year according to the source. Their first public appearance is expected in Las Vegas this month, at the AWS re:Invent conference, where attendees will be able to interact with them. The source notes that the characters and technology could in the future come to Star Wars: Galaxy's Edge and other Disney attractions.
How the interaction works
Unlike the recently debuted Olaf robot, there is no one secretly controlling the Anzellans with a Steam Deck. Each animatronic has hidden cameras and microphones that capture guests' movements and verbal interactions; this data is processed by multiple AI models running on AWS cloud servers.
The source describes an internal distinction: one model for rapid decisions generates immediate responses, such as establishing eye contact to simulate attention, while larger models plan longer interactions. It does not specify which specific models are used or how the architecture is organized.
The AI does not improvise: the script remains human
The key point, according to the material, is that movements and verbal responses are not generated on the spot by AI. Every action and every line selected and triggered by the models was written in advance by Disney Imagineers, to maintain consistency with each character, including their personality quirks. The AI thus guides the actions, but the Anzellans never deviate from the script.
This reduces the risk that the behavior appears out of character, but it also limits improvisation: the interactivity lies in the selection and triggering of responses, not in their autonomous creation. The source does not clarify how extensive the written repertoire is or how an unanticipated question is handled.
Several aspects therefore remain to be verified: practical operation during re:Invent, actual response times, the reliability of interactions with multiple guests at once, and any possible use in attractions. At the moment these are prototypes presented in a technology context, not an announced adoption for the parks.
