Vehicle Path Prediction at Intersections Using Fleet Sensor Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for determining the most probable path (MPP) of a vehicle for driver assistance functions are inaccurate due to oversimplification and require user-specific data, leading to inefficiencies and increased costs.

Innovation Solution

A computer-implemented method using machine learning algorithms trained on map attribute data and fleet sensor data to predict MPP, employing Random Forest, Neural Networks, or Support Vector Machines, without requiring user-specific data, and leveraging data from a fleet of vehicles to improve accuracy and reduce implementation complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning algorithms are trained on user-specific data to improve prediction accuracy, then measurement precision improves, but device complexity and data privacy requirements worsen

Engineering Contradiction:
Improveprediction accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent uses fleet-wide sensor data from multiple vehicles as a proxy for individual user behavior patterns. Instead of training on personal data, the system creates a collective model that copies aggregate driving patterns across the fleet, achieving accurate predictions without complex individual data processing

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The machine learning model is trained on universal fleet data that captures general driving patterns applicable to all users. This single universal model serves all vehicles in the fleet, eliminating the need for separate user-specific models and reducing overall system complexity

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Measurement precision

If complex machine learning models are used to improve prediction accuracy, then measurement precision improves, but computational overhead increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidcomputational overhead
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The machine learning model is trained offline using historical fleet data before deployment. This preliminary training action transfers computational burden to the training phase, allowing the deployed model to make predictions with minimal real-time computational overhead

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system optimizes model parameters and data representation to balance accuracy and computational efficiency. By adjusting parameters such as feature selection, model complexity, and data sampling rates, the system achieves accurate predictions with reduced computational requirements

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If fleet sensor data from multiple vehicles is used to improve prediction accuracy, then measurement precision improves, but data processing complexity increases

Engineering Contradiction:
Improveprediction accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system extracts only the necessary features and patterns from fleet sensor data that are relevant to path prediction. By filtering and selecting only critical data elements, the system reduces processing complexity while maintaining prediction accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The fleet data processing is divided into manageable segments: data collection from individual vehicles, aggregation at central servers, feature extraction, model training, and deployment. This segmentation allows complex data processing to be distributed and managed in smaller, more tractable units

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3871911B1System and method for a driver assistance function of a vehicle
Publication Date: 2026.01.28 BAYERISCHE MOTOREN WERKE AG
  • EP3871911B1 patent drawingFigure 1~2
  • EP3871911B1 patent drawingFigure 3
  • EP3871911B1 patent drawingFigure 4

AI summary

The invention relates to a computer implemented method for determining a most probable path (4) of a vehicle (1) for a driver assistance function of the vehicle (1), which driver assistance function requires information of a road ahead of the vehicle (1). The method comprises: receiving map attribute data (5) from of a map unit comprising information of at least one map having at least one map attribute, and fleet sensor data (6), wherein a fleet represents a number of fleet vehicles (10), wherein each fleet vehicle comprises at least one sensor for acquiring position data and/or motion data which for each fleet vehicle (10) are/is transferred to and stored in a fleet sensor data unit; training a machine learning algorithm (14a - 14c) by inputting training data (8b, 8c, 8d) comprising at least part of both the map attribute data (5) and the fleet sensor data (6), wherein the training data (8b - 8d) is related to an intersection category depending on a predetermined number of road segments (2, 2a - 2f) joining at an intersection (5a - 5i); providing to the machine learning algorithm (14a - 14c) input data (9b - 9j) comprising at least one data attribute (15a - 15g, 16a - 16g, 17a - 17g, 18a - 18c) of a road segment (2, 2a - 2f) joining at the intersection (5a - 5i) of the intersection category from which road segment (2, 2a - 2f) the vehicle (1) is approaching the intersection (5a - 5i); generating in response to receiving the input data (9b - 9j) output data of the machine learning algorithm (14a - 14c) comprising a probability (11d - 11n) for a turn at the intersection (5a - 5i) of the vehicle (1) for each road segment (2, 2a - 2f) joining at the intersection (5a - 5i), and; determining, based on the probability (11d - 11n) for a turn at the intersection (5a - 5i) of the vehicle (1) for each road segment (2, 2a - 2f) joining at the intersection (5a - 5i), a most probable turn (12b - 12d) at the intersection (5a - 5i) of the vehicle (1), and, based on the most probable turn (12b - 12d) of the intersection (5a - 5i) of the vehicle (1), the most probable path (4) of the vehicle. Other aspects of the invention are a software program realizing the method and a system for determining the most probable path of the vehicle for the driver assistance function of the vehicle.