Iterative Decision Model Training for Virtual Character Adaptability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In fighting games, general decision models trained using battle data from multiple characters fail to account for unique character characteristics, leading to limited battle winning rates as they cannot adapt specific battle policies for different character types.
Innovation Solution
A method for iteratively training decision models for each virtual character using battle data from other characters, updating and adding new models to their pools until a specific, high-performing decision model is determined, allowing for personalized battle policies based on character strengths and weaknesses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a general decision model is used for all virtual characters, then the model can be applied universally to any character, but the character characteristics are limited and the battle winning rate is reduced
Solution Approach 1:
The patent divides the general decision model into character-specific decision models by segmenting the training process. Each virtual character receives personalized training using its own battle data, creating specialized sub-models that retain universal applicability while adapting to individual character characteristics. This segmentation resolves the contradiction by allowing the system to use specific models for each character rather than forcing a single general model on all characters.
Solution Approach 2:
The patent applies local quality by customizing the decision model training for each specific virtual character. Instead of using a uniform training approach for all characters, the system tailors the training process to each character's unique battle data and characteristics, making the model quality locally optimized for each character while maintaining overall system versatility.
2Adaptability or versatility
If reinforcement learning is used to train a general AI decision model with battle data from multiple characters, then the model can battle against any character, but it cannot adapt specific battle policies for different character types
Solution Approach 1:
The patent segments the reinforcement learning process into character-specific training sessions. Each virtual character undergoes separate training using its own battle data, allowing the system to maintain generalization capability across characters while simultaneously adapting specific battle policies for each character type through dedicated training runs.
Solution Approach 2:
The patent performs preliminary action by pre-training decision models for each character before actual battles. This advance training allows the system to establish character-specific battle policies in advance, so when battles occur, the models are already optimized for each character's unique characteristics rather than requiring real-time adaptation.
3Reliability
If iterative training is performed for each virtual character using their specific battle data, then the battle winning rate is improved, but the training time and computational resources increase
Solution Approach 1:
The patent merges the iterative training processes for multiple virtual characters into a unified training system. By combining the training workflows and utilizing shared computational infrastructure, the system achieves character-specific model optimization while reducing overall training time through parallel processing and resource sharing across multiple character training sessions.
Solution Approach 2:
The patent maintains continuity of useful action by implementing continuous iterative training that runs efficiently over time. The training process is designed to continuously improve models for each character through repeated battles and learning cycles, maximizing the utilization of computational resources while achieving high battle winning rates through sustained optimization rather than single-pass training.
Data Source
AI summary
A decision model training method and apparatus are provided. The method may include: obtaining model pools of virtual characters, the model pools including decision models corresponding to the virtual characters, and the decision models being used for indicating battle policies adopted by the virtual characters in battles; updating and training nth decision models of the virtual characters based on battle data of a battle between the virtual characters in an nth iteration process to obtain n+1th decision models of the virtual characters; adding the n+1th decision models to the model pools of the corresponding virtual characters; and determining, based on an iterative training end condition being satisfied, decision models obtained by the last round of training in the model pools as target decision models of the virtual characters.