2023 Author: Bryan Walter | [email protected]. Last modified: 2023-05-21 22:24
Engineers from Disney Research have created a prototype of a humanoid robotic head with realistic facial expressions. He knows how to identify a person who is interested in him, turn to him and imitate movements typical for a person: blinking eyes, saccades and periodic head lifts during inhalation. The article was published on the Disney Research website.
There is a famous phenomenon in robotics called the Sinister Valley Effect. It lies in the fact that increasing the realism of a humanoid character does not always increase his attractiveness: if he is realistic enough, but still not as much as a living person, an attempt to give him a human appearance, on the contrary, will cause disgust and fear. A feature of the effect is also that this effect is more pronounced for mobile robots than for static humanoid figures. This is due to the fact that people have facial expressions, which are expressed in many small and irregular movements of different parts of the face and body, and robots, as a rule, are deprived of it.
The research arm of The Walt Disney Company has a long history of creating animated characters as the company uses them in its amusement parks. Most of the characters in these parks have a "cartoon" appearance, so their effect of the ominous valley touches little, but the company also makes realistic robots, for example, the heroes of the Pirates of the Caribbean, and they still have this problem. Nevertheless, Disney Research engineers continue to create more and more realistic robotic skeletons that repeat certain features of human behavior.
In the new work, the engineers taught the robot to maintain eye contact with a person and make small facial movements. They used one of the already assembled robots with a torso and head. The engineers used the neck, eyes, eyelids and eyebrows, although the robot can also move its lips and jaw. Additionally, they equipped him with a depth camera on his chest so that he could see the people in front of him and react to them. The authors came up with a role for the robot: this is an old man with poor hearing and vision, who reads a book and is often distracted by people passing by. If a person comes close to him, he either looks at him disapprovingly, or, if he has already seen him, recognizes and reacts in a friendly manner.
When a person appears in the field of view of the camera, the robot creates a kinematic skeleton (a diagram of body parts) for him. Based on the speed and direction of movement, the distance to the person and whether he makes some kind of greeting, for example, waves his hand, the robot determines the level of interest and decides whether he needs to react. The robot also has a habituation mechanism: if it makes eye contact with a person, the algorithm linearly decreases its index of interest over time.
The index of interest has two thresholds: upon reaching the first, the robot looks at the person, turns its eyes and slightly its neck, and if it increases to the second, higher threshold, then the robot completely switches its attention to the person. When the robot looks and switches attention, saccades (rapid involuntary eye movements) are activated, the direction of which randomly varies between three points. This allows you to simulate real-life eye contact, in which people look at both eyes and mouth.
The engineers note that the new features significantly increase the realism of the character, but recognize some limitations. One of them is that in the current version, the robot's eyes are set so that they look in parallel and the focus is set to infinity. On close contact, this creates the impression that the robot is not looking at the person, but through him.
In 2018, a Japanese engineer created a robot with movable eyes, eyelids and eyebrows that can copy human facial expressions in real time.
