Geometry Map Validation Using Machine-Learned Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional SLAM techniques generate geometry maps with defective areas that can lead to inaccurate navigation and safety issues due to the difficulty in verifying and adjusting large amounts of map data.
Innovation Solution
A system that identifies defective areas in geometry maps using a machine learning model, allows users to adjust these areas via bounding boxes, and initiates actions like data cropping and model retraining to improve map accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional SLAM techniques are used to generate geometry maps, then mapping functionality is achieved, but map accuracy deteriorates due to defective areas
Solution Approach 1:
The patent segments the geometry map into multiple regions by identifying defective areas using machine learning classification. Instead of treating the entire map as a single entity, the system divides it into valid and defective regions, allowing targeted correction of only the problematic areas while preserving accurate map data elsewhere.
Solution Approach 2:
The patent implements a feedback mechanism where users can annotate defective areas by providing bounding boxes around erroneous map regions. This user feedback is processed to correct the machine learning model's identification of defective areas, thereby improving map accuracy and navigation safety through iterative refinement.
2Loss of information
If the entire geometry map is transmitted for validation, then complete map data is available, but data transmission cost increases
Solution Approach 1:
The patent extracts only the defective regions from the complete geometry map for validation and correction. By using machine learning to pre-identify problematic areas, the system extracts only those specific regions that require user attention and data transmission, rather than transmitting the entire map, thereby reducing data transmission costs while maintaining completeness of corrected information.
Solution Approach 2:
The patent applies local quality by focusing validation and correction efforts only on defective regions rather than treating the entire map uniformly. This allows the system to concentrate resources on areas that actually need improvement, reducing overall data transmission requirements while ensuring that map accuracy is enhanced where it matters most.
3Productivity
If machine learning model is used to identify defective areas, then validation efficiency is improved, but model accuracy may be insufficient for complex map defects
Solution Approach 1:
The patent uses feedback from user annotations to continuously improve the machine learning model's accuracy. Users provide bounding boxes around defective areas, and this feedback is used to retrain and refine the model, gradually increasing its ability to accurately detect and classify complex map defects while maintaining high validation efficiency.
Solution Approach 2:
The patent applies partial action by having the machine learning model perform initial defect identification, then using user feedback to correct and refine these identifications. This collaborative approach combines the speed of automated detection with the precision of human expertise, achieving both high validation efficiency and accurate defect detection.
Data Source
AI summary
Systems and methods are provided for verification of the mapping output while accounting for the large amount of data associated with a full geometry map. For example, the system may generate the geometry map of an environment where a vehicle is located and automatically identify a defective area of the geometry map. The defective area may, for example, be identified using a machine learning model to detect the defective area with respect to a threshold or confidence value. The system can receive a bounding box from at least one user device that identifies an adjustment to the defective area of the geometry map. Using the bounding box, the system can crop the defective area of the geometry map and initiate an action based on the adjustment to the defective area of the geometry map.