Autonomous Vehicle AI Module Selection for Redundant Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current approaches to highly automated driving using artificial intelligence (AI) face challenges in ensuring safety and reliability due to the complexity and unpredictability of neural networks, which are difficult to verify and require significant computing resources, leading to inefficiencies and increased costs.
Innovation Solution
A method and device for configuring a control system in autonomous vehicles by dynamically selecting and redundantly executing multiple AI modules based on functional and non-functional properties, such as object recognition and computing capacity, to ensure robust and adaptable performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If neural networks are used for autonomous vehicle control functions, then the system can achieve intelligent decision-making and pattern recognition capabilities, but the system becomes extremely complex and difficult to validate for safety
Solution Approach 1:
The patent divides the complex neural network system into multiple independent neural networks, each responsible for specific control functions (acceleration, braking, steering). This segmentation allows individual networks to be validated separately while maintaining overall system intelligence, resolving the contradiction between adaptability and complexity.
Solution Approach 2:
The patent creates redundant copies of neural networks for critical functions. Multiple neural networks are trained and deployed to perform the same control task, enabling validation through comparison and consensus. This copying approach maintains intelligent capabilities while enabling systematic safety validation.
2Reliability
If multiple redundant resources are allocated for failsafety in autonomous driving functions, then system reliability improves, but the discrepancy between costs and benefits increases due to resource inefficiency
Solution Approach 1:
The patent implements dynamic resource allocation where neural networks are activated based on actual driving conditions and requirements. Instead of continuously running all redundant networks, the system dynamically selects and activates only the necessary networks for current operational contexts, improving resource efficiency while maintaining reliability through selective redundancy.
Solution Approach 2:
The patent changes the operational parameters of redundant neural networks by training them with diverse datasets and configurations. This parameter variation allows the system to leverage diversity in network responses for validation purposes, improving reliability through comparative analysis while optimizing resource utilization by activating only necessary networks.
3Productivity
If a single AI module is used for autonomous vehicle control, then computing capacity utilization is efficient, but the system lacks redundancy and robustness for safety-critical functions
Solution Approach 1:
The patent merges multiple independently trained neural networks into a unified control system architecture. The networks are combined through a fusion mechanism that integrates their outputs, achieving both redundancy for reliability and coordinated operation for efficient resource utilization. This merging approach allows the system to leverage collective intelligence while maintaining computational efficiency.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A method, a computer program with instructions, and a device for configuring a control system for a partially autonomous vehicle. In a first step (10), data on the driving situation are acquired. Selection criteria are then determined from the available data (11). Based on the selection criteria, two or more AI modules are selected from a library of AI modules during driving operation (12). Finally, a combined execution of the selected two or more AI modules is initiated by the control system (13).