Reinforcement learning-driven adaptive covariance control for robust automated INS/UWB navigation.

Area of Science:

• Robotics and Autonomous Systems
• Navigation and Control Systems
• Machine Learning for Sensor Fusion

Background:

• Inertial Navigation Systems (INS) and Ultra-Wideband (UWB) integration is crucial for accurate positioning.
• Dynamic and Non-Line-of-Sight (NLOS) environments pose significant challenges to traditional navigation systems.
• Adaptive filtering techniques are needed to handle uncertainties and improve statistical consistency.

Purpose of the Study:

• To develop a reinforcement learning (RL)-based adaptive covariance scaling framework for robust INS/UWB integrated navigation.
• To enhance navigation performance in dynamic and NLOS-prone environments.
• To enable anchor-wise adjustment of UWB measurement noise for improved accuracy and consistency.

Main Methods:

• Formulated covariance tuning as a Partially Observable Markov Decision Process (POMDP).
• Employed a recurrent Proximal Policy Optimization (PPO) algorithm for policy learning.
• Implemented anchor-wise adjustment of UWB measurement noise.

Main Results:

• Achieved a Root Mean Square Error (RMSE) of 0.258m in simulations, outperforming existing methods.
• Demonstrated centimeter-level accuracy (0.036m RMSE) in real-world quadrotor experiments.
• Showcased strong robustness against severe NLOS and anchor dropout conditions.

Conclusions:

• The proposed RL-based adaptive framework significantly improves the robustness and accuracy of INS/UWB integrated navigation.
• The method is effective for resilient intelligent navigation systems operating in challenging environments.
• Anchor-wise noise adjustment provides a robust solution for balancing accuracy and statistical consistency.