Robot Reference Mapping with Semantic Object Exclusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Localization in robotics using reference maps is computationally expensive and prone to errors due to variability in robot operations and misidentification of static and persistent objects.
Innovation Solution
A robot generates a reference map by scanning a region, performing image analysis using semantic segmentation to exclude dynamic and non-persistent objects, and associating a location identifier with the map, allowing for accurate localization by comparing current image data to the map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the robot uses reference maps for localization, then localization capability is achieved, but computational complexity increases and errors occur due to misidentification of objects
Solution Approach 1:
The patent segments the reference map into multiple layers: a geometric map containing only static structural elements (walls, floors, ceilings) and a semantic map containing movable objects. This segmentation allows the localization algorithm to process only the static geometric layer, reducing computational complexity while maintaining localization accuracy by excluding dynamically changing objects that cause misidentification errors.
Solution Approach 2:
The patent extracts and removes dynamic and movable objects from the reference map used for localization. By taking out these problematic elements that cause misidentification, the system creates a purified geometric map containing only stable structural features, thereby reducing computational errors and improving localization reliability without requiring complex real-time object tracking.
2Loss of information
If the reference map includes all objects in the environment, then complete environmental representation is achieved, but localization accuracy decreases due to dynamic and non-persistent objects
Solution Approach 1:
The patent divides the environmental representation into two separate maps: a geometric map retaining complete structural information (walls, floors, ceilings) for localization, and a separate semantic map or object list containing dynamic and movable objects. This segmentation preserves environmental completeness for navigation purposes while ensuring localization accuracy by using only the stable geometric map.
Solution Approach 2:
The patent extracts dynamic and non-persistent objects from the localization reference map while maintaining their presence in a separate semantic representation. This extraction eliminates sources of localization error caused by moving objects like people, furniture, and temporary structures, while the system retains awareness of these objects for collision avoidance and navigation tasks.
Data Source
AI summary
An example robot performs a scan to obtain image data of a given region. The robot performs image analysis on the image data to detect a set of undesirable objects, and generates a reference map that excludes the set of undesirable objects, where the reference map is associated with the location of the robot at the time of the scan.