Murka — НИОКР КБНЦ РАН

A software package for computer-aided design of intelligent decision-making systems based on multi-agent neurocognitive architectures

About the project

The real world in which robots have to function is an uncertain, unstructured dynamic stochastic, partially observable, episodic active environment. In such an environment, unforeseen, difficult or impossible situations arise systematically, in which robots cannot effectively and safely perform their target functionality. The existing methods of processing unstructured data and decision-making by autonomous robots are not effective enough with changing internal and external environmental parameters, which leads to difficulties in modeling systems capable of self-learning and self-organization.

This problem can be solved only if a number of essential conditions are met, which include, first of all, the use of so-called intelligent software agents with a developed cognitive architecture as the conceptual basis for simulation, which provides the processes of extracting knowledge from input data streams in order to autonomously ontologize them using permanent learning.

Intelligent Agent (IA)

an intelligent system based on a multi-agent neurocognitive architecture, which consists of software agents whose behavior is determined by an internal objective function, the implementation of which is carried out due to the ability of agents to interact with each other.

Multi-agent contract

contractual obligations under which interaction takes place; represent an algorithm according to which agents transfer their available energy in exchange for knowledge

Knowledge

products, the conditional part of which defines the initial and final situation

The core

an action that moves an agent from the initial situation to the final one

Energy

A dimensionless quantity is a measure of the agent’s activity in the environment

Valence

the ability of an agent to enter into contractual relationships with certain types of agents

Each agent in the IA has a knowledge base, according to which it operates and enters into multi-agent contracts. The agent’s behavior can be controlled by editing the rules in the knowledge base. You can edit by adding or removing the whole rule or some of its parts, both in the conditional and in the nuclear component.

All the knowledge that is generated by various agents can be combined into an IA, since the intelligent system allows agents to recurse into each other. In order for an IA based on a multi-agent neurocognitive architecture to function successfully, it must combine:

a multimodal pattern recognition system,
a system for understanding and synthesizing statements,
a situational analysis model,
a system for synthesizing active behavior,
an effector control
system, and a learning system.

Each such system is a functional node that is formed by the self-organization of agents of a certain type, and provides a separate stage of intellectual reasoning (cognitive block). Due to the presence of a sensory subsystem (exteroceptors and interoceptors), IA is able to register external and internal parameters of various modalities. The IA is put into operation and, with the help of interactive provision of missing knowledge, the IA’s research behavior is organized, aimed at the automatic formation or completion of the necessary working ontologies.

This problem can be solved only if a number of essential conditions are met, which include, first of all, the use of so-called intelligent software agents with a developed cognitive architecture as the conceptual basis for simulation, which provides the processes of extracting knowledge from input data streams in order to autonomously ontologize them using permanent learning.