Researchers from Rutgers University introduced in their AI model not only an analog of neurons but also astrocytes – cells that regulate their work. This is a new approach to training robots.
Scientists have used a new approach to neuromorphic research. They imitated the structure of the human brain more fully and did not focus on the work of neurons, and also introduced an analog of astrocytes – they regulate the work of the organ and perform a supporting function. Researchers noted that this is the first development that does not consider neurons as the only processing unit in the brain.
Engineers designed computational models that describe what happens inside an astrocyte when it interacts with neurons during receiving and sending pulses. Then they used these models as building blocks for neural astrocytic networks, which were built into neuromorphic chips that could control robots.
“Since astrocytes may play a key role in the activity of the brain, their study and implementation in machine power is a very exciting and useful direction. The main goal of our study was to study how neurons and astrocytes communicate with each other and build algorithms that allow the body to perform many functions”, the scientists noted.
Astrocytes are able to change the neural activity and behavior of robots. This model of impulse transmission by neurons is fundamentally different from the basic learning algorithms that can only change the network structure. Therefore, researchers were among the first to introduce cells capable of teaching more complex behavior.
The team showed the effectiveness of their approach to a system for controlling the walking of a robot with six legs in a controlled environment. The system allowed the robot to move at different speeds and avoid obstacles. A new approach to neuromorphic computing opens up opportunities for the development of AI, the method has already been included in the curriculum of a course in computer engineering at Rutgers University (USA).