Vehicle Steering Angle Prediction Using Segmented Sensor Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing regression models for autonomous steering systems face challenges such as requiring large datasets, complex architectures, high computational costs, difficulty in deployment on embedded devices, and susceptibility to domain shift issues, leading to inaccurate results, especially in adverse conditions like bad weather.

Innovation Solution

A method that involves using a segmentation model to preprocess sensor data, reducing irrelevant information and simplifying it for a regression model, allowing for more efficient and accurate steering angle predictions by focusing on application-specific data, and enabling independent training and deployment of models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a regression model is applied directly to raw sensor data to determine steering angle, then the model can process all available information, but the computational complexity and resource requirements increase significantly

Engineering Contradiction:
Improvesteering angle accuracyVSAvoidmodel architecture complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the raw sensor data processing task into two distinct stages: first, a segmentation model identifies and extracts relevant objects (such as lane markings, road boundaries, or other steering-relevant features) from the sensor data; second, the regression model processes only this segmented, object-focused data to determine the steering angle. This segmentation approach reduces the complexity of information the regression model must handle while maintaining steering angle accuracy.

Inventive Principle:
Principle #1Segmentation

2Reliability

If a complex architecture with joint training is used to improve model performance, then the model can learn from multiple tasks simultaneously, but the training time and computational resources increase significantly

Engineering Contradiction:
Improvemodel performanceVSAvoidtraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the training process into independent stages: the segmentation model is trained first to identify relevant objects, and then the regression model is trained separately using the segmented data as input. This independent training approach eliminates the need for complex joint training architectures while achieving reliable model performance, significantly reducing training time and computational resource requirements.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If a regression model processes all sensor data including irrelevant information, then comprehensive analysis is performed, but the computational effort and processing time increase

Engineering Contradiction:
Improvesteering angle determination accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent extracts only the relevant information needed for steering angle determination by using the segmentation model to identify and isolate specific objects (such as lane markings or road boundaries) from the sensor data. The regression model then processes this extracted, relevant information alone, eliminating computational waste on irrelevant data while maintaining accurate steering angle determination.

Inventive Principle:
Principle #2Taking out (Extraction)

4Measurement precision

If large datasets are used to train the regression model to improve accuracy, then the model performance increases, but the data acquisition and processing requirements become challenging

Engineering Contradiction:
Improveregression model accuracyVSAvoiddataset size
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts the essential features needed for training by using the segmentation model to identify and isolate relevant objects from sensor data. This extracted segmented data serves as focused training data for the regression model, reducing the overall dataset size required while maintaining or improving model accuracy, as the training data contains only the most relevant information for steering angle determination.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4485366A1Vehicle, apparatus, method for training a regression model, and method for processing sensor data
Publication Date: 2025.01.01 ELEKTROBIT AUTOMOTIVE GMBH
  • EP4485366A1 patent drawingFigure 1
  • EP4485366A1 patent drawingFigure 2
  • EP4485366A1 patent drawingFigure 3~4

AI summary

Embodiments of the present disclosure relate to a vehicle, an apparatus, a method for training a regression model, and a method for processing sensor data of a vehicle. The method for processing sensor data obtaining sensor data of an environment and applying a segmentation model to the sensor data to obtain segmented sensor data of an object in the environment. Further, the method comprises applying a regression model to the segmented sensor data to determine a steering angle for the vehicle.