3D Feature Prediction for Autonomous Driving Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of generating training data for deep learning systems in autonomous driving is labor-intensive and inefficient, requiring significant manual effort for data curation and labeling, which limits the performance of machine learning models.
Innovation Solution
A method that uses sensor data from vehicles to create a training dataset by capturing a time series of elements, including image data and odometry information, to generate accurate three-dimensional representations of features like lane lines, allowing for automated creation of training data and improved model performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data curation and labeling is used to create training datasets, then data quality and accuracy can be ensured, but significant time and labor resources are required
Solution Approach 1:
The system performs preliminary action by using a first trained model to generate predictions and identify training examples before the actual training data creation process. This preliminary prediction step pre-screens and pre-processes the data, so that when manual or automated labeling occurs, the work is already partially done, reducing the time and effort needed for complete data curation while maintaining accuracy
Solution Approach 2:
The system introduces an intermediary component - the first trained machine learning model - that acts as a mediator between raw sensor data and the final training dataset. This intermediary model generates predictions that guide the selection and labeling process, bridging the gap between unprocessed data and labeled training examples, thereby reducing manual intervention time while preserving labeling quality
2Reliability
If diverse and accurate training data is collected to improve model performance, then machine learning model accuracy improves, but the complexity of data collection and curation processes increases
Solution Approach 1:
The system implements feedback by using predictions from the first trained model to identify which training examples would be most beneficial for improving the second model. This feedback loop allows the system to focus data collection and curation efforts on specific, high-impact examples rather than attempting to collect all possible data types, reducing collection complexity while maintaining model performance improvements
Solution Approach 2:
The system changes parameters by dynamically adjusting which data characteristics and features are prioritized for collection and labeling based on the predictions and performance gaps identified by the first model. Instead of collecting all possible data uniformly, the system adapts its data collection parameters to focus on the most valuable examples, simplifying the overall data curation process while ensuring model performance gains
Data Source
AI summary
A processor coupled to memory is configured to receive image data based on an image captured by a camera of a vehicle. The image data is used as a basis of an input to a trained machine learning model trained to predict a three-dimensional trajectory of a machine learning feature. The three-dimensional trajectory of the machine learning feature is provided for automatically controlling the vehicle.