UAPF: A UWB Aided Particle Filter Localization For Scenarios with Few Features

Area of Science:

• Robotics
• Simultaneous Localization and Mapping (SLAM)

Background:

• Lidar-based localization struggles with accuracy in feature-poor environments and robot kidnap situations.
• Existing methods lack robust mechanisms for detecting robot kidnapping and compensating for pose errors.

Purpose of the Study:

• To develop an improved Particle Filter localization (UAPF) for enhanced robot kidnap detection and pose error compensation.
• To improve the accuracy and robustness of robot localization in challenging environments.

Main Methods:

• Adaptive Particle Filter (UAPF) utilizing Ultra-wideband (UWB) Jacobian for covariance updates.
• Novel Kidnap Probability (KNP) criterion for robust robot kidnap detection.
• Pose fusion of ranging-based and Particle Filter-based localization for reduced uncertainty.
• Linear regression of ranging data using maximum probability values for improved accuracy.

Main Results:

• UAPF achieved robot kidnap recovery in under 2 seconds.
• Positioning error was less than 0.1 meters in a 40x15m hall.
• Outperformed prevalent lidar-based localization, which took over 90 seconds and converged to incorrect positions.

Conclusions:

• The proposed UAPF significantly enhances robot localization robustness and accuracy, particularly in scenarios with few features and during robot kidnaps.
• UAPF offers a superior alternative to current lidar-based localization methods, providing faster recovery and higher precision.