Patch Model Generation for Defective Autonomous Driving Scenes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing control models for autonomous vehicles and mobile bodies face challenges in adapting to changing traffic conditions and environments, leading to difficulties in quickly improving performance without significant updates or data collection.
Innovation Solution
A model generation method that collects defective scene data from scenes with insufficient performance and generates a patch model using machine learning or rule-based methods to supplement the control model's performance, allowing for targeted improvements without full model updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the control model is updated to improve performance in specific scenes, then the performance in those scenes is improved, but the complexity and time required for model updates increases
Solution Approach 1:
The patent segments the control model into a main model and separate patch models, where each patch model addresses specific defective scenes. This allows targeted improvements without requiring full model updates, reducing update complexity while maintaining performance reliability in critical scenarios.
Solution Approach 2:
The patent applies local quality by creating patch models that specifically address performance deficiencies in particular scenes rather than uniformly updating the entire control model. This enables localized performance improvement with reduced computational overhead and update complexity.
2Adaptability or versatility
If the control model is fully updated to adapt to changing conditions, then the adaptability is improved, but the time and resources required for updates increases
Solution Approach 1:
By dividing the adaptation task into patch models for specific defective scenes, the system can quickly adapt to changing conditions in those scenes without performing time-consuming full model updates, thus improving adaptability while minimizing update time.
Solution Approach 2:
The patent applies partial action by generating patch models only for scenes with insufficient performance rather than updating the entire control model. This selective approach reduces update time and resources while maintaining necessary adaptability for critical scenarios.
3Reliability
If the control model structure is made more complex to handle various circumstances, then the performance in diverse scenes is improved, but the ease of operation and maintenance deteriorates
Solution Approach 1:
The patent segments the complex control model into a simple main model and multiple specialized patch models. This structure improves performance in diverse scenes by allowing each patch model to specialize in specific scenarios, while maintaining ease of operation through modular, independent patch model management.
Solution Approach 2:
By assigning specific scene-handling capabilities to individual patch models rather than distributing complexity throughout the entire model, the system achieves high performance in diverse scenes while maintaining simple, maintainable model structures through localized specialization.
Data Source
AI summary
The model generation device according to one aspect of the present disclosure collects defective scene data related to a scene that is evaluated as insufficient performance during operations of automatic control using a control model of a mobile body, generates a patch model for supplementing a performance of the control model for the scene evaluated as insufficient performance from the collected defective scene data, and outputs the generated patch model.