ADS Perception Development Using Federated Worldview Baselines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Automated Driving Systems (ADS) face challenges in efficient and cost-effective development and verification, particularly in managing large amounts of data and improving performance without significant impact on size, power consumption, and cost, while ensuring safety and security.
Innovation Solution
A method for performance evaluation and updating of the perception-development module in ADS-equipped vehicles, which involves storing perception data, forming a baseline worldview, matching and evaluating new data against this baseline, and updating perception model parameters using an optimization algorithm to minimize a cost function, facilitating federated learning across production vehicles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If field tests and data collection are conducted to improve ADS performance, then system reliability and safety are improved, but development time and costs increase significantly
Solution Approach 1:
The ADS system performs self-testing and self-evaluation by using its own perception module to generate baseline worldviews and comparing them against ground truth data, eliminating the need for extensive external field tests and manual verification processes
Solution Approach 2:
The system implements continuous feedback loops where perception data is collected, evaluated against ground truth, and used to iteratively improve the perception model, enabling rapid development cycles without prolonged field testing
2Measurement precision
If more sensors and processing power are added to improve perception accuracy, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The system creates virtual copies of the physical environment through baseline worldviews generated by the perception module, allowing detailed analysis and model improvement without adding physical sensors or processing hardware
Solution Approach 2:
The system transitions from physical dimension enhancements (more sensors) to data dimension enhancements (comprehensive baseline worldviews and ground truth data), achieving improved perception accuracy through information rather than hardware
3Productivity
If extensive data is collected and processed to improve model performance, then productivity is improved, but energy consumption and computational resources increase
Solution Approach 1:
The system extracts only the essential perception data and baseline worldviews needed for model improvement, filtering out redundant information to reduce computational load and energy consumption while maintaining productivity
Solution Approach 2:
The system performs preliminary processing to create baseline worldviews and ground truth data in advance, organizing data structures beforehand to enable efficient processing during model training without excessive real-time computational requirements
Data Source
Figure 1
Figure 2(a)~2(d)
Figure 3
AI summary
The present invention relates to methods and systems that utilize the production vehicles to develop new perception features related to new sensor hardware as well as new algorithms for existing sensors by using federated learning. To achieve this the production vehicle's own worldview is post-processed and used as a reference, towards which the output of the software (SW) or hardware (HW) under development is compared. Through this comparison a cost function can be calculated and an update of the SW parameters can be locally updated according to this cost function.