Mobile robot and mobile robot control method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for global localization of moving robots are inefficient, particularly in position jumping events, due to high time requirements for feature matching, inaccurate distance calculations, and the need for three-dimensional maps, which also consume excessive storage and do not consider feature rarity or ambient brightness issues.
Innovation Solution
The solution involves dividing the travel area into small and large areas based on feature distributions, using a controller to extract recognition features, generate descriptors, and compute feature distributions, while considering rarity and ambient conditions to efficiently recognize the current position without focal distance information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If three or more features are matched with features on a pre-stored three-dimensional map to recognize current position, then position recognition capability is achieved, but time required for feature matching becomes excessively long
Solution Approach 1:
The patent divides the travel area into multiple small areas and creates separate feature distributions for each small area. When position recognition is needed, the system first identifies which small area the robot is in, then only matches features within that specific small area rather than searching through the entire map. This segmentation dramatically reduces the computational scope and time required for feature matching while maintaining accurate position recognition.
2Loss of information
If a three-dimensional map is used to store travel area information, then comprehensive environmental information is available, but storage requirements become excessive
Solution Approach 1:
The patent extracts only the essential feature information from the three-dimensional map and stores it in a simplified format. Instead of storing complete three-dimensional spatial data, the system extracts key features and their distributions within each small area, creating a compressed representation that retains sufficient information for position recognition while significantly reducing storage requirements.
3Productivity
If features are matched only within a feature cluster on a three-dimensional map, then matching speed is improved, but accuracy of position estimation is reduced
Solution Approach 1:
The patent segments the map into multiple small areas, each with its own feature distribution. This allows the system to search within a limited scope (improving speed) while still considering all relevant features in that area (maintaining accuracy). The segmentation creates an efficient search space that balances both speed and accuracy requirements.
4Device complexity
If focal distance information and two-dimensional coordinates are used to calculate distance to features, then distance calculation is simplified, but accuracy is reduced due to ambient brightness and inclined plane issues
Solution Approach 1:
The patent replaces traditional geometric distance calculation methods (which rely on focal distance and coordinates) with a feature distribution-based matching approach. Instead of calculating distances using optical parameters that are sensitive to ambient brightness and surface inclination, the system compares feature distributions directly, which are more robust to these environmental variations.
Data Source
AI summary
The present invention relates to a moving robot capable of recognizing a position on a map and a control method of the moving robot, and the moving robot according to the present invention includes: a travel drive unit configured to move a main body; an image acquisition unit configured to acquire images of surroundings; and a controller configured to recognize a current position. The controller is further configured to separate the travel area by the predetermined criterion into a plurality of large areas, into which the plurality of small areas is grouped; compute each large area feature distribution and the at least one recognition descriptor by a predetermined superordinate estimation rule to select a large area in which the current position is included; and compute the small area feature distribution and the at least one recognition descriptor by the predetermined estimation rule to select a small area, in which the current position is included, from among a plurality of small areas included in the selected large area.The control method according to the present invention includes: a learning process of learning a travel area to generate a map and separating the travel area into a plurality of small areas by a predetermined criterion; and a recognition process of selecting a current position on the map. The recognition process includes: a recognition descriptor generation process of acquiring an image of the current position, extract at least one recognition feature from the acquired image, and generating a recognition descriptor corresponding to the at least one recognition feature. The recognition process includes: a large area selection process of computing each large area feature distribution and the at least one recognition descriptor by a predetermined superordinate estimation rule to select a large area in which the current position is included.