Mobile robot and control method of mobile robot
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for localizing a mobile robot's position in a global zone are inefficient due to long matching times for feature points, inaccurate estimation caused by camera focal distance issues, high storage requirements for 3D maps, and degradation in estimation accuracy due to descriptor scarcity and unfair similarity calculations.
Innovation Solution
The method involves generating descriptors based on feature points, selecting comparison target nodes, and using label descriptors to rapidly localize the robot's position, even in large learning data scenarios, by classifying nodes into local zones and using global label descriptors to enhance accuracy and reduce storage needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature points are matched between current image and 3D map to localize robot position, then localization accuracy is improved, but matching time becomes excessively long in wide global zones
Solution Approach 1:
The patent divides the global zone into multiple local zones and creates local 3D maps for each zone. When localizing, the system first identifies which local zone the robot is in, then only matches feature points within that specific local map rather than searching the entire global map. This segmentation dramatically reduces the number of feature points to compare while maintaining localization accuracy within the local area.
Solution Approach 2:
The patent pre-divides the global zone into local zones and pre-generates local 3D maps with their respective feature points during the mapping phase. This preliminary organization allows the localization system to quickly identify the relevant local zone and retrieve only the necessary feature points for comparison, avoiding the need to process the entire global map during real-time localization.
2Measurement precision
If 3D map is generated from feature points to enable accurate localization, then localization capability is improved, but storage capacity requirement increases
Solution Approach 1:
Instead of storing a single large 3D map of the entire global zone, the patent segments the environment into multiple smaller local zones, each with its own 3D map. This segmentation reduces the storage burden for each individual map while collectively covering the entire workspace. The system only loads and processes the local map relevant to the current robot position.
Solution Approach 2:
The patent extracts only the essential feature points and their 3D coordinates from the environment to create the local 3D maps, rather than storing complete high-resolution images or detailed geometric models of every surface. This extraction approach retains sufficient information for accurate localization while significantly reducing storage requirements.
3Ease of operation
If camera focal distance information is used to compute distances from feature points, then distance calculation is simplified, but estimation accuracy degrades due to focal distance inaccuracies
Solution Approach 1:
The patent replaces the optical-based distance calculation method (which relies on camera focal distance and 2D coordinate projections) with a direct 3D coordinate comparison method. By using the known 3D coordinates of feature points from the pre-built local map and comparing them with 3D points derived from the current image, the system calculates distances more accurately without being constrained by camera calibration errors or focal distance inaccuracies.
4Measurement precision
If all feature point data on map is used for localization, then localization accuracy is improved, but comparison time increases significantly
Solution Approach 1:
The patent segments the feature point data into multiple local zones, each containing only the feature points relevant to that specific area. During localization, the system identifies the current local zone and compares feature points only within that segment rather than processing all feature points from the entire global zone. This segmentation maintains comprehensive feature point utilization within the relevant area while dramatically improving comparison speed.
Solution Approach 2:
The patent applies local quality by creating specialized local 3D maps for different zones, where each local map contains feature points optimized for that specific area. This allows the system to use high-quality, dense feature point data locally for accurate localization while avoiding the computational burden of processing feature points from distant or irrelevant areas of the global zone.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A method of controlling a mobile robot includes a learning initial operation of acquiring images for respective points, generating descriptors that respectively correspond to a plurality of feature points extracted from the images, and generating nodes that correspond to the images acquired at the respective points, a label generation operation of generating a label descriptor based on the plurality of descriptors, a localization initial operation of acquiring a localization image when a jumping case occurs, and generating respective localization descriptors corresponding to a plurality of localization feature points extracted from the localization image, a comparison target selection operation of matching the label descriptor to each of the localization descriptors and selecting one or more comparison target nodes corresponding to the matched label descriptor, and a last node selection operation of selecting a node estimated as the current position among the one or more comparison target nodes.