Autonomous Self-Learning System

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

This invention describes an autonomous, self-learning control system based on artificial neural networks that incorporates both environmental states and emotional states into its learning and decision-making processes. The core idea is to allow an agent (such as a robot or vehicle) to learn and adapt to new and unknown environments without prior modeling of the surroundings. By integrating representations of emotions (such as pain, pleasure, or hunger) into neural network inputs and outputs, the system develops an intrinsic motivation mechanism. Furthermore, the architecture can simulate and learn both its own behavior and a model of the external world, enabling sophisticated, adaptation-driven control similar to how living organisms learn from experience and emotions.

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

• Autonomous robotics (e.g., self-adapting robots for factories, exploration, or home use)
• Self-driving vehicles that adapt to new, unstructured environments
• Space exploration rovers (such as Mars rovers) that learn terrain and adapt strategies autonomously
• Medical assistant robots adjusting behavior according to patient interaction and changes
• AI agents in complex simulations or virtual environments learning optimal strategies without explicit programming
• Industrial process control systems that autonomously optimize based on feedback and changing conditions

BenefitsContent extracted from patent full text and abstract with AI.

• Eliminates the need for explicit modeling or detailed predefined reward structures—system can learn from sparse emotional cues (such as danger or success).
• Highly adaptive: The system continuously learns and updates its understanding of both self-state and environment, allowing rapid response to novel situations.
• Reduced human involvement in training or supervision, as system leverages internal feedback (emotional representation) for autonomous learning.
• Generalizable across various fields due to the modular, neural architecture and lack of reliance on environment-specific design.
• Potential for safer and more robust operation, as the agent can recognize harmful or suboptimal states (e.g., low battery, collisions) and alter behavior to avoid them.
• Supports modular extension: additional functions (e.g., speech, vision, planning) can be integrated to enhance the agent's capabilities.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

Disclosed according to the invention is a method for controlling a technical system using a first agent (S), wherein: the first agent (S) implements a first artificial neural network (NN1); a first input vector (x) of the first neural network (NN1) and a current state (ht) of the first neural network (NN1) are converted together into a new state (ht+1) of the first neural network (NN1); from the new state (ht+1) of the first neural network (NN1) a first output vector (y) of the first neural network (NN1) is generated; a second input vector (e) representing an emotion is additionally fed to the first agent, said vector being taken into consideration during the conversion of the neural network into the new state; and a second output vector (e') representing an expected emotion of the new state (ht+1) of the first neural network (NN1), is generated.