Utility AI Computational Graph for Debuggable Behavior Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for modeling complex state transitions in virtual gaming and simulation environments, such as Finite State Machines and behavioral trees, struggle with improved decision-making and debugging when states are numerous and hierarchical.
Innovation Solution
A computational graph-based approach using utility AI models, comprising consideration nodes, utility curve nodes, aggregator nodes, and a selector node, to determine appropriate behaviors in virtual environments based on numeric value transformations and aggregations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If Finite State Machines are used to model game/simulation states, then state transitions can be controlled, but the system becomes difficult to debug when states are complex and numerous
Solution Approach 1:
The patent segments the complex state transition system into hierarchical behavior trees with modular nodes (conditional nodes, behavioral nodes, sub-trees). Each node represents a discrete, testable unit of behavior, allowing developers to debug individual nodes independently rather than tracing through complex FSM state transitions. This segmentation maintains reliable state control while dramatically improving debuggability through modular isolation of behaviors.
2Ease of operation
If hierarchical structures are expanded to manage numerous states, then state organization improves, but debugging remains difficult
Solution Approach 1:
The patent implements nested behavior trees where sub-trees are embedded within parent behavioral nodes, creating a hierarchical structure that naturally organizes complex states. Each nested level represents a more specific behavior context, allowing states to be organized from general to specific. The nesting structure maintains clear parent-child relationships that simplify debugging by allowing developers to trace behavior execution flow through defined hierarchical paths rather than managing flat complex state tables.
3Ease of operation
If behavioral trees are used to organize game/simulation tasks, then behavior organization improves, but decision-making capability does not improve due to FSM context constraints
Solution Approach 1:
The patent introduces dynamic utility AI evaluation within behavioral tree nodes, replacing static FSM transition logic. Each behavioral node contains utility functions that dynamically evaluate multiple factors and compute scores to determine behavior selection. This dynamic evaluation allows the system to adapt decision-making based on current game state, agent characteristics, and environmental conditions, significantly enhancing versatility while maintaining the organizational benefits of behavioral trees.
Solution Approach 2:
The patent changes the decision-making parameters from simple state transitions to multi-factor utility evaluations. Each behavioral node considers multiple parameters (utility scores, weights, thresholds, agent attributes, environmental factors) that can be adjusted and optimized. This parameter-based approach allows flexible decision-making where behaviors are selected based on computed utility values rather than fixed FSM transition rules, enabling adaptive and versatile behavior selection while preserving behavioral tree organization.
Data Source
AI summary
A method, a device and a system of training an adaptive utility AI model in a virtual gaming and/or a simulation environment is disclosed. In accordance therewith, the adaptive utility AI model is trained using one or more utility AI algorithms with data solely generated from a bot as an agent interacting with the virtual environment to generate a computational graph. A number of consideration nodes, utility curve nodes, aggregator nodes and a selector node are implemented through the computational graph. The selector node is utilized to select a behavior out of a set of behaviors applicable to the virtual environment as an appropriate behavior of the agent in the virtual environment in accordance with the implementation of the computational graph.