The National Science Museum of Japan has a new exhibit - an android named Alter. It is very similar to the geminoids of the great robotics guru, Professor Hiroshi Ishiguro, but technologically it is a much more complex machine. The fact is that the android works on the basis of a neural network that allows it to move independently. For the movement of the arms and head, 42 pneumatic actuators are used, the control of which is assigned to the central generator of ordered activity.
At the heart of this android model is a digital unit that simulates the operation of neurons, as well as a set of sensors and sensors that collect information about the distance to objects, ambient temperature and humidity. Thanks to this, the robot can independently move its arms and head. Of course, the movements of the robot are far from the smoothness of human movements, but such an unusual set of technologies integrated into this machine, for some reason, make others perceive it as something that is really alive.
This project is an attempt to build a connection between programming the movements of robots and the possibility of their independent movements. However, the current level of neural networks makes the robot's movements too abrupt and, as the researchers say, "chaotic." In fact, it is not surprising, because the movements of the hands and the head are performed by Alter at the discretion of the system.
The neural network responsible for movements has to process many parameters simultaneously and on this basis independently choose among two modes of movements: longer and smoother or chaotic, which we talked about above. The decision about the choice is largely based on the information received from the sensors. They calculate what is happening around the robot and transmit signals to the digital neural unit. Essentially, all of these sensors behave like a robotic version of the skin. They copy the workings of our senses, although, of course, on a much, much more primitive level. For example, if the sensors detect the presence of a large crowd of people near the robot, then the robot's reaction to the environment will be torso movements.
Android Alter can "sing". True, such singing is possible only in nightmares. Imagine a sort of metal singing of sirens (not security systems, but mythical creatures), which is accompanied by the movements of all the moving parts of the car at once.
The very principle of operation of the generator of ordered activity of this android is based on the mathematical model of Izhikevich's neural activity, which works on the principle of "impulse behavior".
Promotional video:
"When something happens nearby, the system creates an impulse signal that propagates along a chain to other neurons."
Professor Ikeyu from the University of Tokyo describes the work of the generator as an "ordered pendulum" in which one ball hits another, which in turn hits the third, fourth, and so on, creating the movement of the entire system as a whole. And while the movements of the robot do not have the same rhythm balance as the pendulum, Alter works in its own rhythm. Here the movements are not given by robotics. The robot makes movements independently.
Osaka University's Kuhei Ogawa, who previously worked on humanoid android models at Hiroshi Ishiguro's lab, comments:
“Alter doesn't look human. He doesn't move like a human. However, the robot definitely carries a sense of presence. It really feels like something alive, as if it is not a robot, but at the same time it is not a person."
“Until now, the principle of the robot's interaction with the outside world was programmed manually. Just imagine what kind of work it is to program an android to adequately interact with the environment for at least 10 minutes. Alter, in turn, moves independently. His movements are not programmed in any way from the human side."
At the Japanese museums in Tokyo and Osaka, the robot will be shown to the public for a week. During this time, scientists hope to get some really interesting ideas from visitors about what else the android Alter can teach.
NIKOLAY KHIZHNYAK
