Vehicle Localization Using Signatured Gaussian Mixture Maps

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing vehicle/robot localization methods face challenges in accurately matching real-time sensor data with pre-mapped environments, particularly in dynamic conditions.

Innovation Solution

The method employs Signatured Gaussian Mixture Models, which combine Gaussian Mixture Models of geometric primitives with signatures and existence probabilities to represent map elements. These models are used to match real-time point clouds or images with pre-mapped Signatured Gaussian Mixture Maps.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional laser point matching with grid map is used for localization, then the system is simple to implement, but localization accuracy deteriorates in dynamic conditions

Engineering Contradiction:
Improvelocalization accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the environment representation into multiple map types (grid map, point cloud map, feature map, NDT map, NDT-OM map) and processes different map elements separately. Each map type handles specific aspects of the environment, allowing the system to maintain simplicity while improving accuracy through specialized processing for each map type.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a composite localization system that integrates multiple map representations and matching algorithms. By combining grid maps, point cloud maps, feature maps, and NDT maps into a unified localization framework, the system achieves higher accuracy in dynamic conditions while managing complexity through modular architecture.

Inventive Principle:
Principle #40Composite materials

2Measurement precision

If detailed point cloud maps are used for localization, then localization accuracy improves, but storage requirements increase

Engineering Contradiction:
Improvelocalization accuracyVSAvoidstorage requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts and stores only the essential features and characteristics needed for localization from complete point cloud data. By storing feature descriptors, geometric primitives, and probabilistic occupancy information rather than full point cloud datasets, the system maintains high localization accuracy while significantly reducing storage requirements.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of storing complete point cloud maps and processing them for localization, the patent inverts the approach by storing pre-extracted features and descriptors that can be quickly matched against sensor data. This inversion reduces storage needs while maintaining or improving localization efficiency.

Inventive Principle:
Principle #13The other way round (Inversion)

3Reliability

If multiple map types are processed for localization, then robustness in dynamic conditions improves, but processing time increases

Engineering Contradiction:
Improverobustness in dynamic conditionsVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent applies partial processing by selectively using different map types based on the specific localization scenario and environmental conditions. Rather than always processing all map types equally, the system dynamically selects the most appropriate map representation, reducing processing time while maintaining robustness when needed.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent changes processing parameters dynamically based on environmental conditions, sensor availability, and localization requirements. By adjusting which map types are processed and at what detail level, the system maintains robustness in dynamic conditions while optimizing processing time for different operational scenarios.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3311110B1Method and apparatus for locating vehicle/robot
Publication Date: 2025.01.15 BAYERISCHE MOTOREN WERKE AG
  • EP3311110B1 patent drawingFigure 1
  • EP3311110B1 patent drawingFigure 2(a)~2(b)
  • EP3311110B1 patent drawingFigure 2(c)~2(d)

AI summary

A method and an apparatus for representing a map element and a method and an apparatus for locating a vehicle/robot based thereupon. The method for representing a map element comprise: generating a Gaussian Mixture Model for the map element (110); generating a signature for identifying the map element, wherein the signature comprises properties of the map element (120); and generating Gaussian Mixture Model comprises the Gaussian Mixture Model, the signature and an existence probability of the map element (130). A novel technology for representing map elements and an improved vehicle/robot localization technology are provided based thereupon.