Machine Controller Configuration Using Interpretable Action Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine controller configurations optimized by data-driven learning methods are often complex and difficult for experts to interpret, hindering validation and further development.
Innovation Solution
A method and system that generate and optimize action execution trees by replacing predefined action patterns with references, using numerical optimization methods like genetic optimization, to create configurations with high performance and low complexity, allowing for easier interpretation and validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data-driven learning methods are used to automatically configure machine controllers, then configuration time is reduced and performance is optimized, but the configurations become incomprehensible and uninterpretable by experts
Solution Approach 1:
The patent introduces behavior trees as an intermediary representation between raw data-driven learning outputs and expert interpretation. Behavior trees provide a structured, hierarchical format that experts can understand while still capturing the optimized control logic, thus mediating between automated optimization and human comprehensibility
Solution Approach 2:
The patent creates simplified copies or abstractions of the complex learned policies through behavior tree structures. Instead of presenting raw neural network weights or complex state-machine transitions, it generates a copied representation in terms of actionable, interpretable behavior nodes that preserve essential control logic while improving understandability
2Productivity
If complex configurations are used to optimize machine performance, then output and efficiency are improved, but validation and further development become difficult
Solution Approach 1:
The patent segments complex control configurations into modular behavior tree nodes representing distinct actions and conditions. This segmentation allows experts to validate individual nodes independently and understand the overall system through compositional reasoning, making validation tractable even for complex machines
Data Source
AI summary
To configure a machine controller by an action execution tree, predefined action patterns are read in. A multiplicity of action execution trees for a machine to be controlled is also generated. For a respectively generated action execution tree, a performance for controlling the machine based on the respective action execution tree is determined. The predefined action patterns are also sought in the respective action execution tree. An action pattern found in the respective action execution tree is then replaced at least in part by a reference to the predefined action pattern. A tree size of the thus modified action execution tree is furthermore determined. Based on the generated action execution trees, a numerical optimization method is then used to determine an action execution tree that is optimized with regard to better performance and smaller tree size, and this is output in order to configure the machine controller.