Driving Policy Determination via Saliency-Based Data Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current autonomous driving systems face challenges in determining effective driving policies due to the need for large amounts of accurate and salient data, which is time-consuming and expensive to obtain, especially in complex driving environments.

Innovation Solution

A method that records vehicle sensor data, extracts driver behavior data, and determines driving policies by correlating vehicle events with driver behavior, enabling the development of models that emulate and improve upon human driving skills, particularly in complex environments, and can be used to train autonomous vehicle control systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If large quantities of accurate and salient data are collected through manual techniques, then training data quality improves, but time and expense increase significantly

Engineering Contradiction:
Improvedata accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system automatically determines data saliency using machine learning models that analyze sensor data patterns, eliminating the need for manual human labeling. The computational system self-evaluates which data points are most relevant for training, performing the filtering function that would otherwise require human experts.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Manual human techniques for data filtering and saliency determination are replaced with automated computational algorithms. The system uses machine learning models to automatically identify and prioritize salient training data, substituting the mechanical process of human review with an automated digital system.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If manual filtering and labeling techniques are used to determine data saliency, then data quality improves, but expense increases

Engineering Contradiction:
Improvedata saliency determinationVSAvoidtraining cost
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The computational system automatically performs data filtering and saliency determination using embedded machine learning algorithms, eliminating the need for expensive human expert intervention. The system serves itself by autonomously identifying which training data points are most valuable.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces expensive manual human techniques with automated computational methods. Machine learning models process and evaluate data saliency, substituting the costly human expertise with an automated digital system that scales more efficiently.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If conventional training methods are used, then system complexity remains manageable, but autonomous driving safety and effectiveness decrease

Engineering Contradiction:
Improveautonomous driving safetyVSAvoidtraining system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary layer of machine learning models that bridge raw sensor data and driving policy determination. These intermediate processing layers analyze patterns in sensor data and generate saliency scores, creating a structured approach to handling complex data while improving safety outcomes.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The training system is segmented into distinct functional components: sensor data collection, pattern recognition models, saliency determination algorithms, and policy generation modules. This segmentation allows each component to be optimized independently while working together to improve overall system reliability.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3759700B1Method for determining driving policy
Publication Date: 2023.03.15 NAUTO INC
  • EP3759700B1 patent drawingFigure 1
  • EP3759700B1 patent drawingFigure 2
  • EP3759700B1 patent drawingFigure 3~4

AI summary

Systems and methods for driving. A driving data set for each of plurality of human-driven vehicles is determined. For each driving data set, exterior scene features of an exterior scene of the respective vehicle are extracted from the exterior image data. A driving response model is trained based on the exterior scene features and the vehicle control inputs from the selected driving data sets.