Lanelet Classification Using Simulated Data and Attention Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lanelet classification models for autonomous vehicles require cumbersome and time-consuming manual labeling of training data, which is often unbalanced, with insufficient representation of urban driving scenarios, limiting their effectiveness.
Innovation Solution
A lanelet classification system utilizing a neural network that computes attention scores and normalized shared attention mechanisms to fuse feature vectors, trained with simulated data combined with manual annotations and noise models to mimic real-world variations, including occlusion features, to improve classification accuracy across diverse driving environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If manual labeling of training data is used, then the training data can be obtained, but the process is cumbersome and time-consuming
Solution Approach 1:
The patent uses simulated perception data generated from map data and simulated driving scenarios as a copy or substitute for real-world manually labeled data. This simulated data can be automatically annotated without manual intervention, dramatically reducing labeling time while providing sufficient training examples for the neural network
Solution Approach 2:
The system generates its own training data through simulation environments that automatically create labeled perception data. The simulated data includes automatic annotations of lanelets and road features, eliminating the need for external manual labeling efforts
2Quantity of substance
If training data is collected from real-life driving events, then the data can be used for training, but the data samples are unbalanced and dominated by highway driving scenarios
Solution Approach 1:
The patent employs a simulation environment that can dynamically generate diverse driving scenarios including urban, suburban, and highway conditions. The simulation can adaptively create the specific scenario mix needed for training, ensuring balanced representation of all driving contexts rather than being limited to naturally occurring data distributions
Solution Approach 2:
The patent transitions from collecting data in the real world to generating data in a simulated virtual environment. This dimensional shift allows for controlled generation of underrepresented scenarios (like urban driving) that would be difficult or time-consuming to capture in real-life data collection
3Measurement precision
If a neural network is built to classify lanelets, then classification capability is achieved, but the model requires extensive manually labeled training data
Solution Approach 1:
The patent substitutes simulated perception data with automatic annotations for the extensive manual labeling traditionally required. The simulation generates realistic sensor data with ground truth labels automatically, providing sufficient training material without manual intervention
Solution Approach 2:
The patent introduces a simulation environment as an intermediary between data collection and model training. This intermediary automatically generates both the perception data and its corresponding labels, bridging the gap between raw sensor inputs and labeled training examples
Data Source
AI summary
A lanelet classification system for an autonomous vehicle includes one or more controllers including a classifier having a neural network that classifies lanelets of a lane graph structure based on one or more lane attributes. In one embodiment, the one or more controllers execute instructions to build the neural network. In another embodiment, the one or more controllers train the neural network to classify each lanelet of the lane graph structure.