Modular AI Control System for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Monolithic AI systems for autonomous vehicles are difficult to modify retroactively, especially when traffic rules change, as the unconscious rules generated from training data are not readable for humans and require a full or partial retraining, leading to unintended behavior changes and increased complexity and costs.
Innovation Solution
A control system with separate AI modules for distinct driving maneuvers and object recognition, allowing for modular training and adaptation, reducing hardware costs and enabling selective modifications, with an interface layer for abstraction and normalization of inputs and outputs, facilitating easier adaptation to different vehicle platforms and fault detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a monolithic AI system is used for autonomous driving, then the system can handle complex traffic situations through comprehensive training, but retroactive modifications become significantly more difficult and require full retraining
Solution Approach 1:
The patent divides the monolithic AI system into multiple specialized AI modules, each responsible for specific driving maneuvers (e.g., left turns, right turns, going straight) or object recognition categories. This segmentation allows individual modules to be modified and retrained independently when traffic rules change, avoiding the need for full system retraining while maintaining adaptability.
2Adaptability or versatility
If a monolithic AI system is used, then comprehensive functionality is achieved, but hardware costs and resource requirements increase significantly
Solution Approach 1:
By segmenting the AI system into smaller specialized modules, each module can be optimized for its specific function with appropriate hardware resources. This allows the system to achieve comprehensive functionality while using less total hardware resources compared to a monolithic system, as each module can operate on hardware with smaller dimensions.
3Ease of repair
If AI modules are functionally separated into multiple modules, then selective modification and reuse become possible, but additional training outlay is required for each module
Solution Approach 1:
The patent implements preliminary action by pre-training specialized AI modules during the development phase for specific driving maneuvers and object recognition tasks. This preliminary training creates reusable, independently functional modules that can be quickly deployed and replaced without requiring extensive on-site training, thus reducing the time loss during operational deployment and repair.
4Reliability
If a monolithic AI system is used, then unified control is maintained, but defect localization and repair become more difficult
Solution Approach 1:
The patent segments the AI system into functionally independent modules, each handling specific driving maneuvers or recognition tasks. This segmentation enables precise defect localization by isolating issues to specific modules, making it easier to detect and measure problems. The modular architecture maintains system reliability through coherent control while improving defect detectability and repairability.
Data Source
AI summary
A control system for an autonomous vehicle, including at least one artificial intelligence module, or AI module, which converts input variables into output variables through a parameterized internal processing chain, the parameters of the processing chain being capable of being configured, in a training phase, in such a way that training values of the input variables are converted into the corresponding training values of the output variables, at least one first AI module, which supplies output variables for carrying out a first driving maneuver, and a second AI module, which supplies output variables for carrying out a second driving maneuver, and/or at least one first AI module that is designed to recognize a first object or a first group of objects and a second AI module that is designed to recognize a second object or a second group of objects, being provided.