Cooperative-Platform Relative Navigation and Predictive-Terminal Guidance for Shipboard UAV Landing in GPS-Denied Environments

Abstract

Autonomous unmanned aerial vehicle (UAV) landing on a moving ship deck is challenging even with Global Positioning System (GPS), and in GPS-denied seas, the vehicle must estimate its motion relative to the heaving deck while committing to a touchdown whose location and time are not yet known. This article presents an integrated estimation, guidance, and control architecture that treats the landing vessel not as a passive target but as an active cooperative sensing platform, transmitting time-stamped inertial packets to the UAV over a delayed and lossy wireless link. These packets are fused with the local UAV inertial measurement unit (IMU), ultrawideband (UWB) anchor–tag ranges, and altimetry through a cross-platform covariance-coupled dual extended Kalman filter (CP-DEKF). By preserving the cross-covariance between the fast relative-navigation states and the slow UAV and vessel IMU biases, the filter lets UWB and altimeter updates correct the biases indirectly, although no sensor observes them directly. A finite-horizon identifiability condition and a companion sensitivity analysis establish when this channel is effective and how packet delay and covariance scheduling reduce the correction it provides. For guidance, a predictive-terminal sequential convex programming (PT-SCP) planner linearizes the moving touchdown manifold about the free final time under a quadratic terminal-accuracy bound, and a linear time-varying (LTV) model predictive control (MPC) closes the loop. In simulation, the CP-DEKF outperforms single-platform extended Kalman filter (EKF) baselines, most notably in velocity, with the margin set by vessel-link latency, packet quality, and UWB geometry. A 600-trial Monte Carlo campaign across Sea States 2–5 yields closed-loop landing success rates of 85%–72% within a 0.5 m threshold.