Vehicle Arithmetic Architecture for Deep Learning Safe-Stop Backup
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous driving systems using deep learning for environmental recognition and route calculation fall short of achieving the required functional safety level of ASIL-D, as they typically operate at ASIL-B, necessitating an enhancement in the automotive arithmetic system's safety features.
Innovation Solution
An automotive arithmetic system comprising a main device that uses deep learning for route generation and a backup device that generates a safe stop route, with a selector device prioritizing backup control signals over main control signals in case of failure, ensuring the vehicle can safely stop at a preset position, and an additional unit that recognizes objects using predetermined rules to set safe areas and adjust routes accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning is used for environmental recognition and route calculation, then the autonomous driving function is achieved, but the functional safety level remains at ASIL-B instead of reaching the required ASIL-D
Solution Approach 1:
The arithmetic system is segmented into a main arithmetic device that performs deep learning-based environmental recognition and route calculation, and a backup arithmetic device that provides rule-based safety verification. This segmentation allows the system to achieve ASIL-D safety level by separating the innovative deep learning functionality from the safety-critical verification functionality, thereby resolving the contradiction between achieving autonomous driving and meeting safety requirements.
Solution Approach 2:
The backup arithmetic device acts as an intermediary that verifies the safety of routes generated by the main arithmetic device. It checks whether generated routes satisfy predetermined safety rules and constraints, serving as a mediator between the deep learning-based route generation and the final route execution, ensuring ASIL-D safety level while maintaining the benefits of deep learning.
2Reliability
If a backup arithmetic device is added to improve safety, then the functional safety level increases, but the system complexity and computational load increase
Solution Approach 1:
The system segments arithmetic functions into two distinct devices: the main arithmetic device handles complex deep learning computations for environmental recognition and route generation, while the backup arithmetic device handles simpler rule-based safety verification. This segmentation improves safety without requiring the backup device to replicate the full complexity of the main device.
Solution Approach 2:
The backup arithmetic device uses simpler, more reliable rule-based algorithms that are computationally less intensive and easier to verify for safety. Rather than using complex deep learning models in the backup device, it employs straightforward safety rules that are faster to execute and easier to certify, effectively using simpler computational objects to ensure safety.
3Productivity
If the system uses only deep learning for route calculation, then the route optimization is improved, but the safety assurance for emergency stopping is insufficient
Solution Approach 1:
The backup arithmetic device serves as an intermediary safety layer that specifically verifies emergency stopping requirements and safety constraints for routes generated by the main arithmetic device. It checks whether routes satisfy predetermined safety rules including emergency stopping capabilities, thereby ensuring safety assurance without compromising the route optimization efficiency of the deep learning-based main device.
Solution Approach 2:
The backup arithmetic device performs preliminary safety verification before routes are executed, checking whether deep learning-generated routes satisfy predetermined safety rules and constraints. This preliminary anti-action prevents unsafe routes from being executed, ensuring emergency safety assurance while allowing the main device to focus on optimized route calculation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An automotive arithmetic system (SY) includes: a main arithmetic device (100) that determines a target motion of a motor vehicle so that the motor vehicle travel on a route generated based on a vehicle external environment estimated using deep learning based on an output from a vehicle external information acquisition device (M1); and a backup arithmetic device (300) that determines a backup target motion for causing the motor vehicle to travel on a travel route which is generated based on the output from the vehicle external information acquisition device (M1) and which the traveling motor vehicle takes until the motor vehicle stops at a stop position that satisfies a preset criterion. The automotive arithmetic system (SY) outputs a backup control signal to actuators in preference to a control signal when the main arithmetic device (100) fails.