Semantic-Assisted Robot Localization for Laser SLAM Loop Closure

Solution Overview

Problem

Current 2D laser SLAM algorithms for robot localization suffer from mis-matching due to limited information in laser point clouds, leading to high false detection rates and inaccurate loop closure detection.

Innovation Solution

A localization method utilizing semantic point clouds and global semantic grid maps to identify candidate poses, combining coarse localization with semantic information and fine localization using laser point clouds to reduce computation and improve accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If 2D laser SLAM algorithms are used for robot localization, then the robot can perform loop closure detection and relocalization, but the limited information in laser point clouds causes mis-matching and high false detection rates

Engineering Contradiction:

Improveloop closure detection accuracyVSAvoidinformation in laser point clouds

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent combines laser point cloud data with semantic point cloud data to create a more informative representation of the environment. By merging these different data sources, the system overcomes the information limitation of laser point clouds alone, enabling more accurate loop closure detection and reducing false detection rates through multi-source data fusion.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces semantic information as an intermediary element that bridges the gap between laser point cloud data and meaningful environmental understanding. Semantic point clouds serve as a mediator that adds interpretive layers to raw laser data, enabling the system to distinguish between different types of objects and structures, thereby reducing mis-matching in loop closure detection.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If semantic point clouds and global semantic grid maps are used for coarse localization, then localization accuracy is improved, but computation complexity increases

Engineering Contradiction:

Improvelocalization accuracyVSAvoidcomputation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the localization process into two distinct stages: coarse localization using semantic point clouds and global semantic grid maps, followed by fine localization using laser point clouds. This segmentation allows the system to first establish a rough position estimate using computationally efficient semantic matching, then refine it with more precise but computationally intensive laser point cloud processing, thereby managing overall computation complexity while maintaining high accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary coarse localization using semantic information before conducting fine localization with laser point clouds. This preliminary action narrows down the search space and provides an initial pose estimate, which significantly reduces the computational burden of the subsequent fine localization step, as the system only needs to search within a limited region around the coarse estimate rather than the entire map.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If laser point cloud matching is performed for fine localization, then localization precision is improved, but the amount of computation required increases

Engineering Contradiction:

Improvelocalization precisionVSAvoidcomputation required

Core Design Contradiction:

Measurement precisionVSPower

Solution Approach 1:

The patent performs preliminary coarse localization using semantic information before conducting fine localization with laser point clouds. This preliminary action narrows down the search space and provides an initial pose estimate, which significantly reduces the computational burden of the subsequent fine localization step, as the system only needs to search within a limited region around the coarse estimate rather than the entire map.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent segments the localization process into two distinct stages: coarse localization using semantic point clouds and global semantic grid maps, followed by fine localization using laser point clouds. This segmentation allows the system to first establish a rough position estimate using computationally efficient semantic matching, then refine it with more precise but computationally intensive laser point cloud processing, thereby managing overall computation complexity while maintaining high accuracy.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12430797B2Localization method and apparatus, electronic device, and storage medium

Publication Date: 2025.09.30 BEIJING YOUZHUJU NETWORK TECH CO LTD

AI summary

The present disclosure relates to a localization method and apparatus, an electronic device, and a storage medium. The localization method comprises: acquiring a target local semantic point cloud map, a global semantic grid map, and a laser point cloud map; performing pose identification on the basis of the target local semantic point cloud map and the global semantic grid map to obtain a set of candidate poses; and determining a target pose according to the laser point cloud map and the set of candidate poses.