LiDAR-IMU Sensor Fusion for Real-Time Adaptive Vibration Suppression in Autonomous Passenger Boarding Bridges: Full-Scale Field Validation

Abstract

Passenger boarding bridges are increasingly adopting autonomous docking to improve airport operational efficiency; however, contact-induced structural vibrations, whose modal characteristics vary with aircraft class, cabin height, and docking speed, continue to limit safety and docking accuracy. Conventional input-shaping and fixed-gain controllers operate reactively using only proprioceptive measurements and cannot cope with such time-varying dynamics. This paper presents the design, real-time realization, and full-scale field validation of a preview-aware adaptive input-shaping framework that couples exteroceptive three-dimensional light detection and ranging with an inertial measurement unit under a model predictive control horizon. A dual-band sensing architecture is formulated from the Nyquist characteristics of each sensor: low-frequency structural modes between 0.5 and 5 Hz are estimated from point cloud-based motion tracking of the bridge-aircraft contact geometry, while high-frequency modes between 5 and 50 Hz are captured by the onboard inertial sensor. The estimated modal parameters continuously reconfigure a dual-stage zero-vibration-derivative shaper whose impulse sequence is embedded as a soft preview constraint within the predictive cost, simultaneously enforcing trajectory tracking, vibration suppression, and actuation limits. An input-to-state stability argument, detailed in the Appendix, establishes closed-loop boundedness under bounded frequency-estimation error. The framework is implemented on a full-scale 12.5-ton boarding bridge at Incheon International Airport and evaluated docking trials spanning eight operational scenarios, including sensor-degradation cases. Experimental results demonstrate a 73.5 % reduction in residual vibration amplitude, a 75.4 % reduction in overshoot, and a 63.4 % improvement in settling time against five conventional baselines, while maintaining real-time feasibility at an average cycle time of 18 milliseconds.