DESIGN TECHNOLOGY OF THE ARTIFICIAL INTELLIGENCE THAT SUPPORTS AUTONOMOUS DISTRIBUTED SYSTEMS

Abstract

The need for more advanced autonomous systems is confirmed by the growing demand not only from the military domain but also from the civilian community. The efforts of developers of autonomous systems based on Artificial Intelligence (AI) are focused on increasing the level of autonomy of future systems. Known AI models cannot comprehensively solve the problem of autonomy due to the gap between the two paradigms of "processing data from sensors" and "making decisions based on expert knowledge in verbal form." The paper discusses a new model of AI, known as Feeling AI (FAI), which is a hybrid AI based on cognitive models borrowed from living beings. Since the FAI’s Knowledge Base (KB) is represented by a set of independent universal Knowledge Granules (KG) that interact according to the event model, the Event-Driven Architecture (EDA) is most suitable for implementing the processing in FAI. The primary challenge of using EDA is that its components must process data in real-time when they are received at different times and with varying aging characteristics. FAI employs a novel cognitive model of data aging, which supports the gradual forgetting of events over time and estimates their current relevance using a fuzzy Certainty Factor (CF). The design of an FAI's perception system, based on EDA technology, is considered using the example of servicing an autonomous system of distributed objects, such as the Automatic Garage Complex (AGC). The AGC includes a wheeled robot that receives data on the state of a separate box in addition to the stationary perception system. Separate stages of logical design are presented, starting with expert knowledge, its formalization in the form of a set of KGs of the KB, mapping the structure of the KB into hardware deployment, and representing the knowledge that the Event Broker EDA operates with in the parameter table. At the physical design stage, a software implementation of the FAI core was developed in C++ using the ZeroMQ library to provide high-performance asynchronous messaging. The proposed architecture implements an event dispatching mechanism through router components and computational threads, which allows the system to be scaled by parallelizing the execution of the knowledge granule processing function.