Federated Perception Model Training Using Baseline Worldview
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Automated Driving Systems (ADS) face significant challenges in developing and verifying perception algorithms due to the high cost and time-consuming process of creating hand-labelled datasets, which are essential for training machine learning models, and there is a need for efficient solutions to improve the development and verification of ADS features without increasing the size, power consumption, or cost of the on-board system.
Innovation Solution
A method for weakly annotating data using a computing system that stores sensor data and perception data from vehicle-mounted sensors, forms a baseline worldview, evaluates the data against this baseline, and updates perception models using a weakly supervised learning algorithm, reducing the need for manual annotation and enabling efficient development of new perception features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If hand-labelled datasets are used for training perception algorithms, then training data quality is improved, but development cost and time consumption increase significantly
Solution Approach 1:
The system uses the ADS perception module to automatically generate annotations for training data. The perception module processes sensor data and generates worldview information that serves as training annotations, eliminating the need for manual human annotation. This self-service approach allows the system to create its own training data automatically.
Solution Approach 2:
The patent introduces a learning platform as an intermediary between the ADS perception module and the training data generation process. This platform coordinates the collection of sensor data, generates worldview information through post-processing, and creates training datasets with automatic annotations, mediating the complex process of training data preparation.
2Measurement precision
If hand-labelled datasets are used for training perception algorithms, then training data quality is improved, but development cost increases
Solution Approach 1:
The system uses the ADS perception module to automatically generate annotations for training data. The perception module processes sensor data and generates worldview information that serves as training annotations, eliminating the need for manual human annotation. This self-service approach allows the system to create its own training data automatically.
Solution Approach 2:
The patent introduces a learning platform as an intermediary between the ADS perception module and the training data generation process. This platform coordinates the collection of sensor data, generates worldview information through post-processing, and creates training datasets with automatic annotations, mediating the complex process of training data preparation.
3Productivity
If more sensors and processing power are added to improve perception algorithm development, then data processing capability is improved, but system size and cost increase
Solution Approach 1:
The learning platform utilizes existing ADS components (sensors, perception module, processing units) for multiple purposes. The same sensor data used for real-time driving decisions is also used for generating training data and evaluating perception algorithms. This multi-functionality eliminates the need for separate dedicated hardware for development purposes.
Solution Approach 2:
The system uses the ADS perception module to automatically generate annotations for training data. The perception module processes sensor data and generates worldview information that serves as training annotations, eliminating the need for manual human annotation. This self-service approach allows the system to create its own training data automatically.
Data Source
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. In case of a large discrepancy between the baseline worldview and perceived worldview by the module-under-test, the data is weakly annotated by the baseline worldview. Such weakly annotated data may subsequently be used to update the SW parameters of the “perception model” in the module-under-test in each individual vehicle, or to be transmitted to the “back-office” for off-board processing or more accurate annotations.