Autonomous Self-Learning System

Simple SummaryContent extracted from patent full text and abstract with AI.

The invention describes an autonomous self-learning system that uses interconnected artificial neural networks (ANNs) to control technical systems (like robots or autonomous vehicles). The system can adapt to new, unknown environments without requiring detailed external models or supervised training data. This is achieved by incorporating both a 'world model' (representing the system’s environment) and an 'emotion model' (representing intrinsic motivators, such as pain or reward) as part of the learning process, enabling the agent to self-evaluate and adapt its behavior based on outcomes and internal states.

Use CasesContent extracted from patent full text and abstract with AI.

• Control of autonomous robots that need to operate in diverse or changing environments.
• Operation of autonomous vehicles (such as self-driving cars) that must adapt to unexpected situations without explicit programming for each possible scenario.
• Development of exploration robots (e.g., for planetary missions) that can learn about and navigate unknown terrains autonomously.
• Smart medical diagnostic or robotic systems able to learn and adjust to individual patient responses.
• Adaptive industrial automation systems that self-optimize for changing production conditions.
• Artificial agents in interactive simulations and advanced gaming AI that evolve behaviors without pre-coded rules.

BenefitsContent extracted from patent full text and abstract with AI.

• Enables autonomous adaptation to unknown or dynamic environments without pre-programmed models.
• Reduces the need for large, annotated datasets or detailed environmental modeling for training.
• Allows more natural, flexible, and robust learning akin to biological systems by integrating intrinsic motivators or emotions.
• Improves efficiency and reliability of autonomous systems in real-world unpredictable scenarios.
• Supports modular extension (e.g., adding language, memory, or planning modules) for more complex tasks.
• Facilitates the development of highly generalizable artificial intelligence applicable across multiple domains.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

Disclosed is a method for controlling a technical system using a first neural network (NN1) of an agent (S), wherein: a first input vector (x) and a current state (ht) of the first network (NN1) are converted together into a new state (ht+1) of the first network (NN1), from which state a first output vector (y) of the first network (NN1) is generated; the first output vector (y) of the first network (NN1) is fed to a second neural network (NN2), a first output vector (x') of the second network (NN2), said vector representing an expected reaction of the second network (NN2) to the first output vector (y) of the first network (NN1), being generated from the new state (wt+1) of the second network (NN2); and the first output vector (x') of the second network (NN2) is compared to the first input vector (x) of the first network (NN1), in order to train the first network (NN1).