Distributed Graph SLAM Partitioning for Scalable HD Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing single-node solvers for graph SLAM do not scale well with the size of the mapping region, making it difficult to build high-definition (HD) maps for larger areas.
Innovation Solution
A distributed computing system using the Apache Spark framework and consensus ADMM algorithm to partition the graph into subgraphs, allowing for scalable and efficient computation by minimizing edge span across subgraphs and utilizing a preconditioning matrix for faster convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single-node solver is used for Graph SLAM, then the algorithm is simple to implement, but it does not scale well with the size of the mapping region
Solution Approach 1:
The patent partitions the global graph into multiple subgraphs that can be processed in parallel across different nodes. Each subgraph contains a subset of vertices and edges, allowing distributed computation while maintaining the overall graph structure. This segmentation enables the system to scale to large mapping regions by distributing the computational load across multiple nodes.
2Productivity
If the graph is partitioned into multiple subgraphs, then scalability is improved, but the number of edges spanning different subgraphs increases computation complexity
Solution Approach 1:
The patent assigns different roles (constrained, non-constrained, native) to vertices based on their local properties within each subgraph. This local differentiation allows the algorithm to handle boundary vertices differently from internal vertices, optimizing the treatment of edges that span subgraphs while maintaining overall consistency through the ADMM framework.
3Productivity
If a distributed algorithm is used, then the system can handle large-scale HD map building, but convergence speed may be reduced without proper preconditioning
Solution Approach 1:
The patent introduces a positive-definite preconditioning matrix E_k that modifies the Hessian matrix to improve the condition number of the optimization problem. This parameter transformation accelerates convergence by making the optimization landscape more favorable, allowing the distributed ADMM algorithm to converge faster despite the complexity of handling large-scale partitioned graphs.
Data Source
Figure 1(a)~2
AI summary
A method of solving a graph simultaneous localization and mapping (graph SLAM) for HD maps using a computing system includes partitioning a graph into a plurality of subgraphs, each of the subgraphs having all of the vertices of the graph and a subset of the edges of the graph. Constrained and non-constrained vertices are defined for the subgraphs. An alternating direction method of multipliers (ADMM) formulation for Graph SLAM is defined using the partitioned graph. A distributed Graph SLAM algorithm is then defined in terms of the constrained and non-constrained vertices based on the ADMM formulation. The distributed Graph SLAM algorithm is then used to solve the Graph SLAM problem for HD maps.