A Kalman Filter-Based Localization Calibration Method Optimized by Reinforcement Learning and Information Matrix...

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Dynamic environments pose challenges for Unmanned Aerial Vehicle (UAV) swarm localization due to filter parameter degradation and inefficient data fusion.
• Traditional localization methods struggle with adaptability and error propagation in complex, changing conditions.

Purpose of the Study:

• To develop an optimized Kalman filter-based localization calibration method for UAV swarms in dynamic environments.
• To enhance localization accuracy, robustness, and system consistency through adaptive parameter adjustment and advanced data fusion.

Main Methods:

• A Kalman filter-based localization calibration method integrating reinforcement learning (RL) and information matrix fusion (IMKF) was proposed.
• An actor-critic RL network adaptively adjusts the state covariance matrix for improved Kalman filter adaptability.
• A multi-trajectory information matrix fusion strategy aggregates trajectory data in the information domain to minimize error propagation.

Main Results:

• The proposed RL-IMKF method demonstrated superior localization accuracy and stability compared to traditional Extended Kalman Filter (EKF) methods.
• Experimental results using simulated and real-world sensor data validated the effectiveness of the RL-IMKF approach.
• The method significantly improved cooperative localization calibration for UAV swarms in dynamic scenarios.

Conclusions:

• The RL-IMKF method offers a robust and adaptive solution for UAV swarm localization in challenging dynamic environments.
• This research provides a significant advancement in cooperative localization, enhancing the reliability and performance of multi-UAV systems.
• The adaptive nature of the RL-enhanced Kalman filter and the fusion strategy are key to overcoming limitations of existing methods.