Semantic Map Production System with Entropy-Based User Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semantic map production systems for robots face challenges in accuracy and user-friendliness, particularly in indoor environments, as they require human intervention for map modification and may inaccurately recognize meanings based on object recognition technology, necessitating a system that can update maps probabilistically and interactively with users.
Innovation Solution
A semantic map production system that utilizes a 3D image and RGB camera to create metric and spatial semantic maps, employing probability-based methods for object classification, Bayesian updates, and user interaction to refine uncertain areas through entropy-based questioning and semantic networks, allowing non-expert users to correct and update maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If object recognition technology and ontology are used to recognize meanings in a topological map, then the map can be created with extracted information on nodes and edges, but the meanings may be inaccurately recognized
Solution Approach 1:
The system implements feedback through user interaction where users can provide corrections to the robot about object identities and locations. The robot processes this feedback through the map correction unit, which updates the semantic map accordingly. This closed-loop feedback mechanism allows the system to improve recognition accuracy over time while maintaining the simplicity of topological map representation.
2Measurement precision
If a grid map is used to express the environment, then location estimation can be performed with metric information, but the map requires large memory and complex processing
Solution Approach 1:
The system segments the environment representation into two distinct components: a topological map for global structure and navigation routing, and a semantic map with probabilistic object information for meaning recognition. This segmentation allows location estimation to be performed using simplified topological relationships rather than complex grid-based metric processing, reducing computational complexity while maintaining functional capability.
3Measurement precision
If expert users modify semantic maps to increase accuracy, then the accuracy of the semantic map improves, but non-expert users cannot easily modify the maps
Solution Approach 1:
The system implements self-service through automated probabilistic updates and entropy-based questioning. When the robot encounters uncertain object classifications, it automatically generates questions for users based on entropy calculations, presenting only the most uncertain cases. This self-service mechanism allows non-expert users to easily improve map accuracy by simply answering straightforward questions without needing to understand complex map modification procedures.
4Extent of automation
If the robot creates a map without user interaction, then the system operates autonomously, but the map accuracy remains limited by object recognition technology
Solution Approach 1:
The system applies partial automation by maintaining autonomous operation for most map creation tasks while selectively introducing user interaction only when entropy thresholds indicate high uncertainty. The map correction unit monitors uncertainty levels and triggers user questions only for specific uncertain cases rather than requiring continuous user interaction, thus preserving system autonomy while improving accuracy where needed.
Data Source
AI summary
The system includes a metric map creation unit configured to create a metric map using first image data received from a 3D sensor, an image processing unit configured to recognize an object by creating and classifying a point cloud using second image data received from an RGB camera; a probability-based map production unit configured to create an object location map and a spatial semantic map in a probabilistic expression method using a processing result of the image processing unit, a question creation unit configured to extract a portion of high uncertainty about an object class from a produced map on the basis of entropy and ask a user about the portion, and a map update unit configured to receive a response from the user and update a probability distribution for spatial information according to a change in probability distribution for classification of the object.