Vehicle Pose Graph Fusion for Real-Time Position Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional positioning techniques using sensor fusion, such as those based on Kalman filters, face limitations in flexibility and precision, particularly in transitioning between different positioning systems and handling variable combinations of position data, leading to unreliable or imprecise estimates of a vehicle's position, especially in dynamic environments.
Innovation Solution
A method employing a pose graph with chain geometry that optimizes odometry and absolute position data from multiple sources, allowing for flexible sensor fusion and efficient computation by using a block tridiagonal system matrix, which enables real-time processing and precise position estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional sensor fusion techniques (e.g., Kalman filters) are used to combine position data from multiple positioning systems, then the system can provide position estimation, but the flexibility to handle variable combinations of position data and transitions between different positioning systems is limited
Solution Approach 1:
The patent segments the position data from multiple positioning systems into discrete data sources, each represented as a separate node in the pose graph. This allows the system to independently evaluate and combine data from different positioning systems (e.g., GPS, visual odometry, inertial sensors) without requiring a unified complex model, thereby improving flexibility in handling variable combinations of position data while maintaining reliability through graph optimization.
Solution Approach 2:
The pose graph structure enables dynamic adaptation by allowing the system to add, remove, or weight different data sources based on their current reliability and availability. The graph optimization process dynamically adjusts the fusion of position data from multiple sources, enabling smooth transitions between different positioning systems while maintaining continuous and reliable position estimation.
2Measurement precision
If complex model formation is performed to create comprehensive position estimates from multiple trajectories and sensor data, then measurement precision can be improved, but computational time increases significantly making real-time use impossible
Solution Approach 1:
The patent divides the complex positioning problem into smaller, manageable segments represented as nodes and edges in a pose graph. Each node represents a specific position estimate from a particular data source, and edges represent the relationships between these estimates. This segmentation allows the system to process position data in discrete, computationally efficient units while maintaining high precision through graph optimization, enabling real-time performance.
Solution Approach 2:
The patent changes the computational parameters by using graph optimization techniques that are more efficient than traditional complex model formation. The pose graph representation allows for optimized computation of position estimates by leveraging the graph structure to reduce computational complexity, thereby achieving both high precision and real-time processing capability.
3Measurement precision
If traditional sensor fusion methods are used to combine odometry and absolute position data, then position estimation can be provided, but the precision and reliability are insufficient in dynamic environments with transitioning positioning systems
Solution Approach 1:
The pose graph structure serves as a universal framework that can handle multiple types of positioning systems (GPS, visual odometry, inertial sensors, etc.) in a unified manner. The graph optimization process universally applies to combine data from different sources regardless of their specific characteristics, thereby improving precision while managing complexity through a single, versatile approach rather than separate handling for each sensor type.
Data Source
AI summary
Absolute position data (605) of a machine are determined for respective multiple times (t.1, t.3, t.5, t.8, t.11), and odometry position data of the machine are also determined. A pose graph (661) is generated, wherein edges (672) of the pose graph (661) correspond to the odometry position data, and nodes (671) of the pose graph (661) correspond to the absolute position data (605). The pose graph (661) is optimized to obtain an estimated position. Optionally, the odometry can also be estimated. A driver assistance functionality of the machine, for example a motor vehicle, can be controlled optionally on the basis of the estimated position. For example, the driver assistance functionality can relate to autonomous driving.