Drones Land Safely on Moving Ships in Rough Seas

Researchers are tackling the significant challenge of enabling autonomous Uncrewed Aerial Vehicle (UAV) landings on marine platforms affected by unpredictable ocean conditions. Haichao Liu, Yufeng Hu, and Shuang Wang, all from The Hong Kong University of Science and Technology (Guangzhou), alongside Kangjun Guo, Jun Ma, and Jinni Zhou, also of The Hong Kong University of Science and Technology (Guangzhou), present a novel framework, SpecFuse, designed to improve landing precision in the face of wave-induced oscillations and wind disturbances. This work is particularly important because current methods often fail to accurately predict platform motion, hindering safe and reliable UAV operation at sea. SpecFuse uniquely integrates frequency-domain wave decomposition with time-domain state estimation, offering high-precision 6-DoF motion forecasting and demonstrably outperforming existing techniques in both simulated and real-world lake experiments, paving the way for critical applications like search and rescue and environmental monitoring.

A new control system enables drones to land reliably on ships moving in rough seas. The technology addresses a long-standing problem of accurately predicting a vessel’s motion, vital for automated deliveries and safety-critical operations. Successful trials demonstrate a substantial improvement over existing methods, paving the way for wider use of drones in challenging maritime environments.

Scientists have developed a new control framework enabling unmanned aerial vehicles (UAVs) to land autonomously on moving marine platforms despite challenging sea conditions. Existing methods often fail to adequately model the specific characteristics of ocean waves, leading to inaccurate predictions and unsuccessful landings. This framework integrates frequency-domain wave decomposition and time-domain recursive state estimation to achieve high-precision, six-degree-of-freedom motion forecasting of uncrewed surface vehicles (USVs).

By explicitly modelling dominant wave harmonics, the system significantly reduces prediction phase lags, refining its understanding of the USV’s movement in real time using data from inertial measurement units (IMUs) without requiring complex calibration procedures. Once the USV’s motion is predicted, a hierarchical control architecture plans dynamic trajectories, accounting for non-convex constraints, and executes them using a learning-augmented predictive controller.

Extensive testing, encompassing 2,000 simulations and eight lake experiments, demonstrates the effectiveness of this approach. Results show a prediction error of 3.2cm, with a corresponding landing deviation of 4.46cm. The system achieved success rates of 98.7% in simulations and 87.5% in real-world trials, all while maintaining a low latency of 82ms on embedded hardware.

These performance metrics represent a 44% to 48% improvement in accuracy compared to current state-of-the-art methods. Beyond improved accuracy, the framework’s resilience to combined wave and wind disturbances positions it as a valuable asset for critical maritime applications. These include search and rescue operations, offshore equipment inspection, and environmental monitoring, where reliable UAV landing is essential for extending mission duration and enhancing operational capabilities. To promote wider adoption and further research, all code, experimental configurations, and datasets will be made publicly available as open-source resources.

Autonomous UAV landing achieves centimetre-level precision on a moving marine platform

Prediction errors reached a mean of 3.2cm across all trials, representing a substantial improvement in forecasting marine platform motion. Landing deviation was measured at 4.46cm, indicating the precision of the UAV’s descent and contact. Success rates in simulations were 98.7%, demonstrating reliability under controlled conditions, while real-world lake experiments yielded an 87.5% success rate, validating performance in more complex environments.

Latency on the embedded hardware was consistently 82ms, vital for real-time control and swift reaction to changing conditions. The system outperformed existing state-of-the-art methods by 44% to 48% in prediction accuracy, suggesting a significant leap forward in autonomous landing on moving platforms. At its core, the work integrates frequency-domain wave decomposition with time-domain recursive state estimation.

The system’s ability to handle wave-wind coupling disturbances is particularly noteworthy. By explicitly modelling dominant wave harmonics, the research effectively mitigates phase lags in motion prediction. Now, the framework refines predictions in real time using IMU data, eliminating the need for complex calibration procedures. The hierarchical control architecture features a sampling-based HPO-RRT* algorithm for dynamic trajectory planning under non-convex constraints, and a learning-augmented predictive controller for disturbance compensation.

By fusing data-driven insights with optimisation-based execution, the controller ensures precise descent. Since the system can simultaneously mitigate phase lags, maintain real-time performance, and resist wind-wave coupling, it offers a complete solution for challenging maritime operations. All experimental configurations and datasets will be released as open-source, promoting reproducibility and further research.

Frequency domain decomposition and hierarchical predictive control for USV motion planning

A spectral-temporal fusion predictive control framework, termed SpecFuse, began with frequency-domain wave decomposition to explicitly model dominant wave harmonics. This aimed to reduce prediction phase lags, a common problem in forecasting the motion of uncrewed surface vehicles (USVs) subject to sea states. By analysing the frequency content of waves, the system could better anticipate future platform movements.

Following decomposition, time-domain recursive state estimation refined these predictions in real time, incorporating data from inertial measurement units (IMUs) without requiring complex calibration procedures. Once motion was forecast, a hierarchical control architecture was implemented. Dynamic trajectory planning utilised a sampling-based hierarchical planning with rapidly-exploring random trees (HPO-RRT) algorithm, designed to navigate the non-convex constraints imposed by the oscillating platform.

This algorithm efficiently searched for feasible paths, accounting for the unpredictable movements of the USV. Subsequently, a learning-augmented predictive controller fused data-driven disturbance compensation with optimisation-based execution, improving landing precision and stability. Extensive validation involved both 2,000 simulations and eight lake experiments, providing a thorough assessment of the system’s performance under varied conditions.

The choice of both simulated and real-world testing ensured the framework’s applicability beyond controlled environments. Furthermore, all code, experimental configurations, and datasets are being released as open-source material, promoting reproducibility and further development within the research community. This commitment to transparency allows other researchers to build upon this work and adapt it to different maritime applications.

Wave frequency modelling enables precise drone landings on maritime vessels

Scientists have devised a system allowing drones to land reliably on moving ships, a feat previously hampered by the unpredictable nature of the sea. For years, automated landings have relied on predicting platform motion, but accurately forecasting wave behaviour, especially the complex interplay of frequencies and wind, proved a stubborn obstacle. Existing approaches often treated ocean swells as random disturbances, or lacked the detail needed to anticipate shifts quickly enough for a safe touchdown.

This new work bypasses those limitations by directly modelling the dominant frequencies within waves, creating a more precise, real-time prediction of platform movement. Achieving this level of accuracy isn’t simply about better forecasting; the team integrated this spectral analysis with a sophisticated control system, enabling the drone to dynamically adjust its approach even as the ship pitches and rolls.

Simulations and lake tests demonstrate a substantial improvement in landing success, exceeding previous methods by nearly half. Beyond the impressive numbers, this represents a shift towards proactive, rather than reactive, control in challenging environments. Reliance on inertial measurement units (IMUs) introduces potential drift over extended periods, and the system’s performance in extreme weather remains an open question.

While lake tests provide valuable data, the open ocean presents a far more complex scenario with larger, irregular waves and stronger winds. The real test will be scaling this technology for practical applications like search and rescue operations or offshore infrastructure inspection. Future work might explore incorporating data from multiple sensors, including cameras and radar, to create an even more comprehensive understanding of the surrounding environment. Ultimately, this advance isn’t just about landing drones; it’s about unlocking the potential of autonomous systems to operate reliably in one of the most demanding environments on Earth.