AI-based adaptive game strategy execution method and AI-based game analysis system

Environment-Adaptive Game Strategy Execution Technology Based on Deep Reinforcement Learning with Imperfect Models

This technology is an AI-based, environment-adaptive game strategy execution method and system that generates real-time policies to address environmental uncertainty by integrating imperfect models from virtual environments with data collected from real-world game environments using deep reinforcement learning.

Existing AI game robots have struggled with discrepancies between virtual and real environments, such as variations in friction, leading to performance errors when executing strategies in real-world settings due to a lack of robustness against uncertainty.

This technology proposes a method that builds an imperfect model reflecting uncertainty factors within a virtual environment and implements a reinforcement learning framework through sequential performance error detection and error function optimization to derive adaptive policies using real-time data feedback from the actual environment. It can be applied to sports robots and industrial precision robots, serving as a core technology to bridge the gap between simulation and reality.

‍

‍

Key Features:

• Providing a virtual environment that includes an imperfect model with extracted environmental uncertainty factors for each type of sport
• Generating environmental changes via the imperfect model within the virtual environment as the game progresses
• Executing reinforcement learning to detect sequential performance errors and optimize error functions and weights
• Deriving environment-adaptive policies through real-time data feedback from the actual environment

‍

This invention was developed with support from the Ministry of Science and ICT for the development of AI curling robot technology capable of establishing game strategies and executing gameplay.