In the past few years, scientists and engineers have developed robots for automated systems such as performing repetitive tasks. Meanwhile, Disney Research has been developing human-like robots with abilities ranging from performing stunts to having eerie eye interactions.
Disney Research recently published a paper that described a realistic and interactive gaze with the Audio-Animatronic humanoid. Previous robot developments have focused on technical implementation with human interaction. The team’s latest advancements include creating a gaze interaction “through the lens of character animation where the fidelity and believability of motion are paramount,” wrote the authors.
For nearly three decades, Disney has been developing animatronic figures, or life-like robots combined with audio and visual elements. These animal or human characters are seen in Disney theme parks around the world.
Realistic Eye Movement
In the video, the robot plays a specific character – of an elderly man reading a book. Despite the decline of his hearing and eyesight, he is constantly distracted by people passing by or greeting him. He takes frequent glances at the movement of people or gives a disapproving stare when people enter his personal space. He also gives a look of recognition towards the people he knows.
Using subsumption architecture, the technology used with AI and robots since the 1980s, the researchers layered motion movements like breathing and blinking. Mechanisms were also designed to mimic human attention behaviors such as responding to stimuli.
Disney’s focus is similar to the concept of animations more than understanding the biology and psychology of eye movement. Focusing on the visual appeal of realistic eye movement is what makes the robot’s gaze convincing and eerie. “The core purpose of animatronics figures is to deliver a show, in the same manner, that actors perform in a theatre,” wrote the authors.
The system also included a curiosity score for every person that becomes visible in its line of sight. The robot will then choose among four behavior states as a response, which is to read, glance, engage, or acknowledge.
The default state of the robot’s eye movement is to read, such as appearing to read a book or newspaper. The glance state is when the attention engine indicates a stimulus in its light of sight and will transition to the read state or engage state.
The engage state is when the attention engine is triggered by the read or glance state, gazing a person with a curiosity score with both eye and head movement. Lastly, the acknowledge state comes after the robot is familiar with the person of interest.
Aside from the behavior selection engine, there is also a database of people, or guestbase, that are part of the scene and make up a type of short-term memory. The guestbase contains three-dimensional points of a person, records of the persons-of-interests for familiarity, and special properties of each guest.
In conclusion, developing realistic gaze interactions with the humanoid is “as an attempt toa scend from the uncanny valley through [the] layering of interactive kinematic behaviors and sensorimotor responsiveness.”