Neural Network Aided Kalman Filtering with Model Predictive Control Enables Robot-Assisted Drone Recovery on a Wavy Surface

TL;DR

The paper addresses robust drone recovery on a wave-disturbed surface by fusing an AI-augmented state estimator with a confidence-aware planner. It introduces KalmanNet++ to predict the UAV’s near-future pose in the manipulator frame and pairs it with a Receding Horizon Model Predictive Controller (RHMPC) that adapts to sea-state variability, torque constraints, and joint limits. The approach yields high reliability (over 95% success) and improvements in efficiency and end-effector precision (up to ~10% and ~20%, respectively) in both simulations and real sea trials, including Beaufort-level disturbances. These results demonstrate the practicality and resilience of cooperative UAV–manipulator systems for maritime operations, enabling safer and more efficient drone recovery and payload handling in challenging marine environments.

Abstract

Recovering a drone on a disturbed water surface remains a significant challenge in maritime robotics. In this paper, we propose a unified framework for robot-assisted drone recovery on a wavy surface that addresses two major tasks: Firstly, accurate prediction of a moving drone's position under wave-induced disturbances using KalmanNet Plus Plus (KalmanNet++), a Neural Network Aided Kalman Filtering we proposed. Secondly, effective motion planning using the desired position we got for a manipulator via Receding Horizon Model Predictive Control (RHMPC). Specifically, we compared multiple prediction methods and proposed KalmanNet Plus Plus to predict the position of the UAV, thereby obtaining the desired position. The KalmanNet++ predicts the drone's future position 0.1\,s ahead, while the manipulator plans a capture trajectory in real time, thus overcoming not only wave-induced base motions but also limited constraints such as torque constraints and joint constraints. For the system design, we provide a collaborative system, comprising a manipulator subsystem and a UAV subsystem, enables drone lifting and drone recovery. Simulation and real-world experiments using wave-disturbed motion data demonstrate that our approach achieves a high success rate - above 95\% and outperforms conventional baseline methods by up to 10\% in efficiency and 20\% in precision. The results underscore the feasibility and robustness of our system, which achieves state-of-the-art performance and offers a practical solution for maritime drone operations.

Figures (20)

• Figure 1: Overview of the cooperative system. (a) The cooperative recovery system consists of a magnetic capture mechanism mounted on a manipulator, which intercepts a tethered UAV. (b) The core challenge: a hovering UAV (global fixed target) is perceived as a rapidly moving object from the drifting base frame of the USV. (c) This relative motion, compounded by non-stationary and coupled base dynamics, introduces substantial uncertainties that complicate accurate pose prediction. (d) Our predictive control strategy estimates the future UAV pose to plan a robust capture trajectory, enabling reliable docking between the magnetic capstan catcher and the UAV's docking block assembly with its Iron plate base.
• Figure 2: EKF block diagram.
• Figure 3: KalmanNet++ main flow. The classical initialisation–prediction–update loop is retained, while the Kalman–gain block is delegated to a learnable Mamba network that is modulated by a hypernetwork conditioned on a scalar state-of-world index $\gamma_t$.
• Figure 4: Architecture 1 --- monolithic Mamba: a single selective SSM layer receives the feature vector $\mathbf{z}_t$ and directly outputs the gain matrix $\widehat{\mathbf K}_t$.
• Figure 5: Architecture 2 --- cascaded Mamba: three lightweight Mamba cells emulate the propagation of the process covariance, the predicted error covariance, and the innovation covariance, respectively. The resulting statistics are fused to form $\widehat{\mathbf K}_t$.