Autonomous Driving Hazard Detection via Multi-Model Redundancy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving systems face challenges in detecting and preventing unsafe control operations due to the limitations of existing redundancy methods, which may not adequately cover systematic and random faults, leading to potential hazardous situations.
Innovation Solution
The implementation of a redundant system using multiple machine learning objects, such as deep neural networks (DNNs), including PlanningNet, SafetyNet, LegalNet, and MoralNet, which operate from different perspectives to verify and modify action planning, providing diverse fault coverage and ensuring safety, legality, and ethical considerations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional redundancy methods are used for fault detection, then systematic and random fault coverage is provided, but the detection capability for hazardous driving situations is insufficient
Solution Approach 1:
The patent applies inversion by training one machine learning model (SafetyNet) to detect unsafe driving situations by learning from inverted or adversarial examples of hazardous behavior. This allows the system to identify hazards by understanding what constitutes unsafe patterns, rather than only learning safe driving patterns, thereby improving hazard detection capability without proportionally increasing system complexity
Solution Approach 2:
The patent introduces an intermediary machine learning-based safety checker that acts as a mediator between the primary driving control system and the actuators. This intermediary layer analyzes sensor inputs and proposed actions to detect potential hazards before execution, providing enhanced fault detection capability while maintaining a modular architecture that limits overall system complexity
2Reliability
If multiple machine learning objects with different perspectives are implemented, then diverse fault coverage and safety are improved, but computational complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-training multiple machine learning models with different perspectives (safety, legality, ethics) during an offline phase. During real-time operation, these pre-trained models can quickly evaluate hazards without requiring extensive computation, thereby providing diverse fault coverage while minimizing processing time loss
Solution Approach 2:
The patent segments the safety checking function into multiple specialized machine learning objects, each responsible for a specific aspect (safety, legality, ethics). This segmentation allows each model to be optimized for its specific task and enables parallel processing of different safety dimensions, improving overall safety coverage while managing computational complexity through modular evaluation
3Productivity
If machine learning objects are trained using the same data, then training efficiency is improved, but detection diversity and hazard identification capability are reduced
Solution Approach 1:
The patent applies local quality by using the same training dataset but applying different preprocessing, augmentation, or selection strategies to create specialized training subsets for each machine learning model. For example, SafetyNet may receive augmented versions of the data with hazardous scenarios emphasized, while LegalNet receives data filtered for traffic rule compliance examples. This allows efficient use of the base dataset while creating diverse detection capabilities across different models
Data Source
AI summary
An autonomous driving system could create or exacerbate a hazardous driving situation due to incorrect machine learning, algorithm design, sensor limitations, environmental conditions or other factors. This technology presents solutions that use machine learning to detect when the autonomous driving system is in this state e.g., erratic or reckless driving and other behavior, in order to take remedial action to prevent a hazard such as a collision.