This article is reproduced by the authorization of the 3D Vision Heart public number, please contact the source.

UWB large-scale real-time industrial positioning

Ultra-wideband (UWB) is a short-range, energy-efficient radio communication technology used for accurate positioning detection and relative distance measurement. For example, devices such as Apple's AirTags and Android's SmartTags are growing in popularity due to their ability to accurately track household items.

While UWB has excelled in applications such as motion tracking, such as football matches, it still faces significant challenges in large-scale real-time industrial positioning applications. Currently, most UWB studies are limited to smaller or indoor Settings and run in absolute positioning (AP) mode, as shown in Figure 1. Under these conditions, all base stations must be calibrated, and labels need to be continuously measured from multiple base stations. In occluded environments, however, this becomes impractical. Such limitations reduce the usefulness of UWB in large-scale areas such as ports or warehouses, where obstacles significantly increase the technical difficulty.

The main challenges of using UWB in large-scale outdoor environments include the frequent occurrence of non-line-of-sight (NLoS) problems and interference from other devices, which can limit the effective measurement range. In the worst case, the connection between base stations may be completely lost due to the addition of a new container (as shown in Figure 1). The lack of reliable GPS signals further complicates calibration. Even after the calibration is completed, the system is still difficult to locate in non-line-of-sight conditions, becoming a bottleneck for field robotics and logistics automation.

To address these issues, this paper presents a comprehensive approach [1] to address the challenges of large-scale UWB calibration and single location. A continuous time LiDAR inertial odometer (CT-LIO) is used to generate trajectories that can be sampled at any point. By interpolating this trajectory, we align each UWB measurement with its corresponding position. Then, Gaussian process is used to iteratively optimize base station location estimation. Although the standard UWB technology is still unable to locate in non-line-of-sight mode after calibration, to solve this problem, a fast descriptor-based approach is combined to enhance the single global positioning capability in large-scale repetitive environments. Our approach significantly improves existing localization methods and introduces new possibilities that have not been explored before.

Major contributions:

1. A method combining UWB data, CT-LIO and Gaussian process is proposed to accurately calibrate UWB base stations in large-scale environments with only one sampling. This approach significantly improves positioning performance in occluded and repetitive environments, and is particularly effective in scenarios where neither UWB nor LiDAR can be used alone.
2. A single location process is proposed. By filtering the UWB range with existing descriptors, the search time is reduced, so that the efficient and accurate location in the complex area is realized, and multiple UWB base stations are no longer required, which makes the scheme economical and practical.
3. The method was verified by real-world experiments to achieve higher accuracy and shorter processing time in a 600x450 square meter environment, demonstrating its practicality for large-scale applications under challenging conditions.
4. The data set and calibration code will be made public for use by the community.

Problem description

Continuous time LiDAR- inertial odometer

The objective function is minimized by a manifold based optimization solver that relies only on the Eigen library for maximum computational efficiency. This optimization method ensures that continuous time tracks are generated within the specified time window, while dealing with noise and deviation problems in the sensor input.

Interpolated UWB pose

Calibration of UWB base station based on Gaussian process

Covariance function

In order to accurately capture spatial relationships, we use the Matern 3/2 kernel function, which can adapt to different levels of smoothness in spatial data:

Among them:

•  Is the smoothness parameter;
•  Is the length scale parameter;
•  It's a modified Bessel function;
•  It's a gamma function.

These parameters are adjusted according to the complexity of the environment and the variability of the data to improve the adaptability of the model.

Prediction equation

Calibration procedure

The specific process of calibration base station location is as follows:

1. Initial fitting: The initial point is first fitted on a coarser scale.
2. Area sampling: Samples undefined locations at regular intervals within a rectangular area.
3. Model update: Select the average of the top 10 best values in the sample value to update the Gaussian process model.
4. Hierarchical random sampling: Ensures that the data covers the entire region, which improves generalization and accelerates convergence.

advantage

Using Gaussian process to calibrate UWB base station location, higher accuracy and robustness can be achieved. Compared with traditional methods, this method can better deal with the problem of base station calibration in complex and non-line-of-sight (NLoS) environments, thus significantly improving the overall positioning performance.

Single fix

We have improved the Stable Triangle Descriptor (STD) to preferentially match scenes near known UWB base stations. This approach aims to reduce mismatches in large-scale duplicate environments by integrating calibrated UWB base station locations with STD descriptors.

Improved STD method

STD relies only on triangular descriptors to identify scenes, but is error-prone in repeating scenes. By combining STD with calibrated UWB data, we introduce range-based partitioning search to enhance location performance.

Single location process

With the improved STD approach, we have implemented the following single-location processes:

1. Preprocessing: According to the calibrated UWB base station location, the search area is divided and the prior descriptors are extracted.
2. Real-time matching: Matches the current STD descriptor with descriptors in the predefined region.
3. Position inference: Quickly locate the current position of the robot or target object based on matching results.

Parameter selection

advantage

Compared to traditional STD methods, the improved solution combined with UWB base station location has the following advantages:

• Improved accuracy: Calibrated base station locations can significantly reduce mismatches in repeated scenarios.
• Enhanced robustness: Maintains high positioning success rates even in non-line-of-sight (NLoS) and complex environments.
• Optimization efficiency: Range-based partition search reduces the search space and thus the computation time.

With this approach, we are able to achieve efficient and accurate single positioning in large-scale repetitive environments, providing a cost-effective solution for industrial logistics and robotics applications.

Experimental effect

Sum up

This paper presents a UWB calibration method for non-line-of-sight (NLoS) problems, using SLICT to generate continuous time trajectories. By sampling the UWB acquired attitude and applying an iterative Gaussian process, we succeeded in achieving a calibration accuracy of about 2 meters in a large-scale environment of 600x450 square meters. Even under non-line-of-sight conditions where traditional UWB positioning fails, the calibrated base station position can be used as a standalone plug-in to improve repeatability and single location success in large-scale environments. Experiments have shown that automated mobile robots for container transport equipped with sparse UWB and on-board LiDAR can achieve accurate single positioning in challenging repetitive environments with minimal cost.