Multi-Sensor Fusion Simultaneous Localization Mapping Based on Deep Reinforcement Learning and Multi-Model Adaptive...

Area of Science:

• Robotics and Artificial Intelligence
• Sensor Fusion and Localization

Background:

• Accurate mobile robot localization is critical for autonomous navigation and simultaneous localization and mapping (SLAM).
• Traditional methods like LiDAR-based Point-to-Line Iterative Closest Point (PLICP) and RGB-D camera-based ORBSLAM2 have limitations in complex environments.

Purpose of the Study:

• To develop a novel multi-sensor fusion technique for robust and accurate mobile robot localization.
• To enhance localization stability and precision in challenging indoor environments using advanced AI.

Main Methods:

• Designed a multi-sensor fusion technique integrating deep reinforcement learning (DRL) and multi-model adaptive estimation (MMAE).
• Employed LiDAR-PLICP and RGB-D ORBSLAM2 for initial localization estimates.
• Utilized Proximal Policy Optimization (PPO)-based DRL with residual value anomaly detection for optimal sensor weight adjustment.

Main Results:

• The proposed method effectively fuses localization information from multiple sensors (LiDAR and RGB-D camera).
• Achieved higher localization accuracy compared to standalone PLICP and ORBSLAM2 methods.
• Demonstrated increased localization stability for mobile robots operating in complex indoor simulation environments.

Conclusions:

• The DRL-MMAE fusion technique significantly enhances mobile robot localization performance.
• The developed approach offers a robust solution for accurate and stable navigation in dynamic environments.
• This work contributes to advancing autonomous systems requiring precise real-time positioning.