Sensor Sub-Region Matching for Low-Complexity Spatial Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing SLAM methods are computationally and memory-intensive, and existing methods for localization and mapping in unfamiliar environments, and the use of SLAM algorithms in autonomous vehicles are computationally expensive and require significant improvements in accuracy, robustness, and computational efficiency, and memory-intensive, limiting their deployment on certain platforms, and the use of SLAM algorithms in autonomous vehicles is costly and requires efficient localization and mapping systems that enable an agent to understand the structure and identity of a space and the objects within it, which is less error-prone and less computationally and memory-intensive.
Innovation Solution
A computer-implemented method for parsing an environment in a multi-dimensional space by obtaining and comparing sub-regions of sensor data to determine relative rotations and generate an action vector for navigation, using reduced-size image data and similarity measures to reduce computational and memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bundle adjustment is used to achieve accurate localization and mapping, then measurement precision is improved, but computational complexity increases significantly
Solution Approach 1:
The patent segments the environment into discrete places and positions, representing the map as a graph structure where places are nodes and transitions are edges. This segmentation allows the system to avoid computationally expensive continuous bundle adjustment by working with discrete, pre-segmented environmental representations that can be efficiently compared and matched.
Solution Approach 2:
The patent performs preliminary segmentation and place recognition before detailed localization. By pre-processing the environment into place segments and storing them in a database, the system eliminates the need for real-time bundle adjustment, as the heavy computational work of environmental parsing is done in advance rather than during navigation.
2Measurement precision
If feature extraction is performed to improve localization accuracy, then measurement precision is improved, but use of energy increases
Solution Approach 1:
The patent extracts only the essential place segments and their key characteristics from the environment, storing them in a database. This extraction approach avoids the energy-intensive continuous feature extraction and bundle adjustment by capturing only the necessary environmental signatures once, then reusing them for multiple localization operations.
Solution Approach 2:
The patent creates copies of place segments and stores them in a database for rapid comparison. Instead of performing expensive feature extraction and bundle adjustment in real-time, the system compares sensor data against pre-stored place copies, dramatically reducing computational energy requirements while maintaining localization accuracy.
3Reliability
If outlier rejection is implemented to improve reliability, then reliability is improved, but computational complexity increases
Solution Approach 1:
The patent performs preliminary place recognition and segmentation before attempting localization. By pre-identifying valid places and their characteristics, the system establishes constraints that automatically filter out invalid matches, eliminating the need for separate outlier rejection steps and their associated computational overhead.
Solution Approach 2:
The patent introduces place segments as an intermediary representation between sensor data and localization results. These place segments act as mediators that naturally filter invalid matches through their inherent geometric and topological constraints, reducing the need for explicit outlier rejection algorithms.
4Reliability
If visual-inertial odometry is used to compensate for convergence failure, then reliability is improved, but measurement precision deteriorates due to accumulated error
Solution Approach 1:
The patent segments the navigation space into discrete places and positions, allowing the system to recognize when a place has been revisited. This segmentation enables direct position updates upon place recognition, eliminating the need for continuous inertial integration and avoiding accumulated position errors that plague VIO-based systems.
Solution Approach 2:
The patent implements feedback through place recognition, where the system continuously compares current sensor data against stored place segments. When a match is found, the system receives feedback about the correct position, allowing it to correct any drift from VIO or other odometry methods and maintain long-term position accuracy.
Data Source
AI summary
A computer implemented method and system are provided for parsing an environment of an agent in a multi-dimensional space, comprising: obtaining first sensor data at a first location of the agent: retrieving stored second sensor data from a second location; obtaining a plurality of first sub-regions of the first sensor data; obtaining a plurality of second sub-regions of the second sensor data; comparing the second sub-region against each first sub-region using a similarity comparison measure to determine a most similar first sub-region to the second sub-region; and determining an associated relative rotation and aggregating the relative rotations for the plurality of second sub-regions to obtain an action vector indicative of an estimated direction from the first location of the agent to the second location.