Robot Positioning via UWB-Visual Joint Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual SLAM systems face challenges with tracking loss and inaccurate positioning due to noise and nonlinearity issues, especially in large-scale environments, leading to insufficient computer resources and errors in map updates.
Innovation Solution
Integration of a UWB device to assist in enhancing the robustness of machine visual positioning by providing absolute distance information, which is synchronized with visual sensor data to improve feature matching and pose estimation, and the use of a joint objective function combining distance and visual residual cost functions for optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If filter-based SLAM methods (EKF-SLAM, PF-SLAM) are used to represent uncertainty of observed information, then positioning robustness is improved, but computer resources are insufficient to update the map in real time after long-term execution
Solution Approach 1:
The patent extracts and removes road signs from the environmental feature map that have been present for a long time. By identifying and eliminating these persistent features, the system reduces the total number of features that need to be tracked and updated, thereby decreasing computational burden and enabling real-time map updates while maintaining positioning robustness through the probabilistic representation of remaining features.
2Measurement precision
If probabilistic methods are used to represent uncertainty in SLAM, then positioning accuracy is improved, but nonlinearity is treated as linearity which leads to errors
Solution Approach 1:
The patent changes the representation parameters by using probabilistic distributions (mean and covariance) to describe road sign positions and uncertainties. This allows the system to properly handle nonlinear transformations through propagation of uncertainty, where the mean and covariance are updated using appropriate mathematical models that account for nonlinearities, thereby maintaining both positioning accuracy and estimation reliability.
3Extent of automation
If visual image feature extraction and matching methods are used to construct environmental feature maps, then simultaneous positioning is achieved, but tracking loss and relocation failure occur
Solution Approach 1:
The patent implements feedback mechanisms by continuously updating the probabilistic representations of road sign positions based on new observations. The system uses the estimated position and uncertainty from previous steps to guide subsequent feature matching and verification, creating a closed-loop system that can detect and correct tracking errors, thereby preventing tracking loss and relocation failure while maintaining autonomous positioning.
Data Source
AI summary
The present disclosure provides a positioning method and a robot using the same. The method includes: obtaining, through the visual sensor, a current frame image; obtaining, through the ultra-wideband tag, distance of a robot from an ultra-wideband anchor; performing a feature matching on the current frame image and an adjacent frame image to generate partial map point(s); determining whether the current frame image is a key frame image; and optimizing a pose of the visual sensor corresponding to the key frame image through a joint objective function in response to the current frame image being the key frame image, where the joint objective function at least comprises a distance cost function of the ultra-wideband anchor and a visual residual cost function. Through the above-mentioned method, the accuracy of the positioning of the robot can be improved.