Neural Network Corrections for Faster VSLAM Map Updates

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

VSLAM technology requires significant computational resources due to the repetitive processing of large amounts of data for bundle adjustment, particularly when new images are added, which affects the efficiency and speed of map information updates.

Innovation Solution

Integrate a neural network inference engine to predict map information updates, reducing the need for extensive computations by using a trained neural network to infer corrections and optimize map information, and adjust feature point distributions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If bundle adjustment is performed repeatedly to update map information, then map accuracy is improved, but computational load and processing time increase significantly

Engineering Contradiction:

Improvemap accuracyVSAvoidprocessing speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system performs preliminary actions by pre-calculating and storing correction amounts for various reprojection error values during training. When updating map information, the system retrieves pre-computed correction amounts from storage based on current reprojection error, avoiding the need for time-consuming iterative bundle adjustment computations during actual map updates.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention creates a computational model (neural network) that copies and generalizes the complex bundle adjustment optimization process. The neural network learns the relationship between reprojection errors and optimal corrections from training data, then applies this learned knowledge to quickly predict corrections for new map updates without performing exhaustive iterative optimization.

Inventive Principle:

Principle #26Copying

2Measurement precision

If iterative bundle adjustment is performed to minimize reprojection error, then map information accuracy is improved, but computational resources and energy consumption increase

Engineering Contradiction:

Improvemap information accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The invention replaces the mechanical iterative optimization process (bundle adjustment using gradient descent and Gauss-Newton algorithms) with a neural network-based prediction system. The neural network substitutes complex mathematical computations with learned pattern recognition, significantly reducing computational energy requirements while maintaining correction accuracy.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes the fundamental parameter of how corrections are computed from iterative numerical optimization to direct neural network inference. By transforming the computation approach from calculating corrections through multiple iterations of bundle adjustment to retrieving pre-computed corrections from neural network storage, the system dramatically reduces energy consumption while preserving map update accuracy.

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If extensive computations are performed for each map update, then correction accuracy is improved, but processing time and system efficiency deteriorate

Engineering Contradiction:

Improvecorrection accuracyVSAvoidupdate processing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs all heavy computational work during the training phase, where neural networks are pre-trained to associate reprojection error patterns with optimal correction amounts. During actual map updates, the system only needs to retrieve pre-computed corrections from storage, reducing update processing time from minutes to milliseconds while maintaining high correction accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The invention prepares and stores correction amount data in advance during training, creating a cushion of pre-computed information that can be quickly retrieved during map updates. This beforehand preparation eliminates the need for time-consuming real-time bundle adjustment computations during actual updates, significantly reducing processing time without sacrificing accuracy.

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

Data Source

PatentUS20250336090A1Map information update method, landmark generation method, and feature point distribution adjustment method

Publication Date: 2025.10.30 SOCIONEXT INC

AI summary

A map information update method includes: (a) obtaining map information; (b) obtaining landmark observed positions indicating positions of one or more landmarks in a captured image; (c) adding that includes (i) generating added map information by adding information pertaining to the landmark observed positions to the map information, and (ii) updating the map information obtained in (a) to the added map information; (d) predicting that includes (i) calculating predicted map information based on the map information updated in (c), by using a neural network inference engine that has been trained, and (ii) updating the map information to the predicted map information; and updating information that includes (i) calculating updated map information based on the map information updated in (d), by using a gradient method, and (ii) updating the map information to the updated map information.