High-Precision Climbing Robot Localization Using Planar Array UWB/GPS/IMU/Barometer Integration

Shuning Zhang, Renjing Xu, Zhanchen Zhu, Xiangyu Chen, Yunheng Wang, Xu Jiang, Peibo DuanarXiv 2025 · DOI · arXiv

Introduction

Climbing robots are increasingly deployed for high-altitude maintenance tasks, such as skyscraper cleaning, wind turbine inspection, and bridge monitoring. However, these vertical environments present a "localization nightmare." Standard positioning tools often fail in these scenarios:

  • GPS/GNSS: Signals are frequently blocked by the building structure (occlusion) or suffer from multipath interference as they reflect off glass and steel surfaces.
  • Vision-based SLAM: Glass curtain walls and uniformly painted steel are often "textureless," providing insufficient visual features for cameras to track.
  • Inertial Sensors (IMU): These suffer from cumulative drift, leading to significant position offsets over time.
  • Magnetometers: Large steel structures cause electromagnetic interference, rendering compass headings unreliable.

To address these challenges, this research proposes a robust, high-precision localization system that fuses multiple sensors using an intelligent attention-based mechanism.

System Overview

The proposed system integrates a planar array Ultra-Wideband (UWB) system with GPS, IMU, and barometers. A key innovation is the use of a planar array UWB anchor. Unlike standard UWB systems that require multiple anchors for trilateration, this planar array can estimate a 3D position (distance plus two angles) from a single anchor point. This significantly simplifies deployment on large structures like skyscrapers.

System overview showing the climbing robot, building environment, and sensor signals.

The Attention Mechanism-based Fusion Algorithm (AMFA)

The core of the system is the Attention Mechanism-based Fusion Algorithm (AMFA). The architecture processes raw sensor data through several stages to ensure high-fidelity positioning:

  1. Deep Learning Correction: Fully Connected Neural Networks (FCNN) are used to process raw UWB and barometer data. These networks are trained to recognize and correct environmental errors, such as signal noise or pressure drift, which traditional geometric or physical models often miss.
  2. Multimodal Attention Mechanism: This serves as the "brain" of the fusion. Instead of simple averaging, it dynamically assigns weights to each sensor based on its current reliability. For instance, if GPS quality (indicated by HDOP) degrades due to occlusion, the system automatically increases its trust in the UWB and barometer data.
  3. Unscented Kalman Filter (UKF): The final fused output is refined through a UKF to ensure the robot's trajectory remains smooth and physically plausible.
The AMFA architecture, illustrating the training and application phases for multi-sensor fusion.

Experimental Results

The researchers evaluated the system using a quadruped climbing robot on the exterior wall of a building, simulating a real-world maintenance scenario. A high-precision RTK (Real-Time Kinematic) GPS base station provided the ground truth for evaluation.

The proposed AMFA system demonstrated significant improvements over traditional methods:

  • High Accuracy: Achieved a Root Mean Square Error (RMSE) of 0.48 m, a substantial improvement over standalone UWB (1.01 m) or barometer (0.89 m) models.
  • Robustness: While standard GPS/INS-EKF approaches saw maximum errors as high as 5.51 m due to signal occlusion, the AMFA system capped maximum error at 1.50 m.
  • Stability: The system consistently stayed closer to the ground truth, outperforming traditional baselines in challenging "semi-occluded" environments.

Conclusion and Future Work

By combining a single-anchor planar UWB array with a deep-learning-based attention mechanism, this research provides a viable solution for high-precision climbing robot localization. While the current system requires a pre-training phase and has a limited UWB range, future work will explore more complex temporal architectures like Transformers and the integration of visual features to further enhance robustness in diverse environments.

Summary updated

BibTeX
@misc{zhang2025high,
  author = {Shuning Zhang and Renjing Xu and Zhanchen Zhu and Xiangyu Chen and Yunheng Wang and Xu Jiang and Peibo Duan},
  title = {{High-Precision Climbing Robot Localization Using Planar Array UWB/GPS/IMU/Barometer Integration}},
  year = {2025},
  eprint = {2509.23801},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.23801},
}
Figure