A Decentralized Sensor Fusion Scheme for Multi Sensorial Fault Resilient Pose Estimation

Area of Science:

• Robotics and Autonomous Systems
• Sensor Fusion
• Navigation and Localization

Background:

• Accurate pose estimation is critical for autonomous systems, but traditional methods struggle with sensor noise and failures.
• Multi-sensor fusion offers improved robustness, yet centralized approaches can be vulnerable to single points of failure.

Purpose of the Study:

• To propose a novel decentralized, two-layered, multi-sensorial fusion architecture for resilient pose estimation.
• To introduce a Fault Resilient Optimal Information Fusion (FR-OIF) paradigm for enhanced accuracy and self-resiliency.
• To validate the proposed scheme's effectiveness and superiority against centralized methods.

Main Methods:

• A two-layered fusion architecture with distributed nodes in the first layer using extended Kalman filters.
• Integration of pose information from diverse sensors (3D lidar, camera, UWB, IMU).
• Implementation of the FR-OIF paradigm in the second layer, employing maximum likelihood fusion and fault isolation.

Main Results:

• The proposed architecture successfully achieved resilient pose estimation for a micro aerial vehicle.
• Experimental results demonstrated high accuracy and robustness in the presence of sensor failures and erroneous measurements.
• The decentralized FR-OIF approach outperformed the classical centralized multi-sensorial fusion method.

Conclusions:

• The novel decentralized two-layered fusion architecture provides a highly resilient and accurate pose estimation scheme.
• The FR-OIF paradigm effectively handles sensor inaccuracies and failures, ensuring reliable localization.
• This approach offers a significant advancement for robust navigation in autonomous systems.