Vehicle Path Probability Modeling at Junction Decision Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing navigation systems struggle with inaccurate path prediction, particularly when drivers deviate from routine routes, as they rely on rule-based calculations that do not account for individual driving habits and are limited by the availability of historical data.
Innovation Solution
A machine-learning model that processes state vectors of traversed and candidate path segments, trained on historic trip data, to predict path probabilities, incorporating both general and user-specific behaviors, using neural networks and contextual information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If rule-based calculations are used to determine path probabilities, then the system can operate without individual historical data, but the prediction accuracy deteriorates because driver habits are not accounted for
Solution Approach 1:
The system dynamically adapts between two operational modes: using rule-based calculations when individual historical data is unavailable, and switching to machine learning model predictions when data exists. This dynamic adjustment resolves the contradiction by allowing the system to maintain operability in data-scarce scenarios while achieving high accuracy when data is available.
Solution Approach 2:
The system changes the operational parameter from pure rule-based calculation to machine learning-based prediction based on data availability. The machine learning model processes state vectors representing driver behavior patterns, transforming the prediction mechanism from static rules to adaptive learned parameters, thereby improving accuracy without sacrificing operability.
2Measurement precision
If individual historical path data is used to calculate probabilities, then prediction accuracy improves for routine journeys, but the system fails for non-routine trips where no historical data exists
Solution Approach 1:
The machine learning model is trained on aggregated historical data from multiple drivers to learn universal driving patterns and behaviors. This universal knowledge enables the model to make accurate predictions for non-routine trips by applying general driving behavior patterns, thus resolving the contradiction between accuracy for routine trips and adaptability for non-routine trips.
Solution Approach 2:
The machine learning model acts as an intermediary between raw historical data and path probability predictions. It processes state vectors that capture driver behavior patterns and transforms them into probability distributions for candidate paths, enabling accurate predictions even when specific individual historical data for a particular trip is unavailable.
3Measurement precision
If machine learning models are trained on extensive historical data to improve accuracy, then prediction precision increases, but computational complexity and data storage requirements increase
Solution Approach 1:
The system extracts only the essential features from historical trip data to create state vectors that represent driver behavior patterns. By focusing on key behavioral features rather than storing and processing all raw historical data, the system achieves high prediction accuracy while reducing computational complexity and storage requirements.
Solution Approach 2:
The machine learning model focuses on learning local driving behavior patterns at decision points rather than attempting to model all possible driving scenarios globally. This localized approach to learning driver habits at critical decision moments improves prediction accuracy while keeping the model computationally efficient and manageable in size.
Data Source
Figure 1~2
Figure 3A~3B
Figure 4
AI summary
The application describes the generation of a model (102) that provides estimates of probabilities for next path segments at decision points (402) (e.g. at junctions) based on (personal) historic traces. The model (102) uses two (or more) road segments (401) that the vehicle has travelled on as inputs to produce the estimates of the probabilities for the next path segments (403,404) for the vehicle.