Autonomous UAV Pipeline Near-proximity Inspection via Disturbance-Aware Predictive Visual Servoing

Pipelines are essential infrastructures for modern energy transportation, and ensuring their structural integrity is critical to prevent leaks, service interruptions, and environmental contamination. Traditional inspection methods rely on manual patrols, which are inefficient, costly, and potentially dangerous, especially in remote areas and complex terrains. In recent years, UAVs have emerged as an attractive alternative due to their mobility and flexible sensing capabilities. However, despite the promise of UAVs in inspection tasks, many existing systems still rely heavily on manual piloting or operator supervision, making it challenging to achieve fully autonomous inspection missions. Researchers have been exploring advanced control strategies and sensing technologies to enhance UAV autonomy and inspection efficiency.

The core problem in UAV autonomous inspection is achieving accurate and reliable near-proximity inspection in complex 3D environments. Traditional GPS-based navigation can become unreliable in signal-degraded or denied environments, and existing visual servoing methods often neglect UAV dynamics, making them difficult to apply to low-inertia and limited-payload quadrotors. Additionally, visual measurements are often available at a lower update rate and with poorer quality than onboard state feedback, further complicating control. Developing a framework that can maintain stability and efficiency under low-rate visual updates, measurement noise, and environmental uncertainties is of significant research value.

The core innovations of this paper include the development of a novel ESKF-PRE-VMPC framework for UAV pipeline inspection tasks. • The framework couples quadrotor dynamics with image feature kinematics, forming a unified predictive model that enables direct image-space prediction within the control loop. • To address low-rate visual updates, measurement noise, and environmental uncertainties, an extended-state Kalman filtering scheme with image feature prediction (ESKF-PRE) is developed, and the estimated lumped disturbances are incorporated into the VMPC prediction model. • Additionally, a terrain-adaptive velocity design is introduced to generate vertical velocity references over unknown terrain slopes. • These innovations enhance the framework's robustness and stability in complex 3D environments.

The methodology of this paper includes several key steps: • Unified Predictive Model: A unified predictive model is formed by coupling quadrotor dynamics with image feature kinematics, enabling direct image-space prediction within the control loop. • Extended-State Kalman Filtering (ESKF-PRE): To address low-rate visual updates, measurement noise, and environmental uncertainties, an extended-state Kalman filtering scheme with image feature prediction is developed, and the estimated lumped disturbances are incorporated into the VMPC prediction model. • Terrain-Adaptive Velocity Design: A terrain-adaptive velocity design is introduced to generate vertical velocity references over unknown terrain slopes, maintaining the desired cruising speed. • High-Fidelity Gazebo Simulations and Real-World Experiments: The framework's effectiveness is validated in high-fidelity Gazebo simulations and real-world experiments.

The experimental design includes validation in high-fidelity Gazebo simulations and real-world environments. • Datasets: Real-world pipeline models are used to simulate various practical scenarios, such as straight and curved segments, uphill and downhill segments. • Baselines: The proposed method is compared against traditional IBVS methods and MPC methods that do not consider quadrotor dynamics. • Metrics: Root Mean Square Error (RMSE) is used to evaluate pipeline orientation and lateral deviation in the image. • Hyperparameters: Key parameters such as prediction horizon length and sampling time used in VMPC are carefully set.

Experimental results show that the proposed ESKF-PRE-VMPC framework reduces RMSE by 52.63% and 75.04% in pipeline orientation and lateral deviation in the image, respectively, for straight-pipeline inspection without wind. Additionally, the method successfully completes both wind-disturbance and bend-pipeline tasks where the baseline method fails. These results demonstrate the framework's improved stability and tracking robustness in complex 3D scenarios compared to existing methods.

The framework can be directly applied to pipeline inspection tasks in the energy transportation sector, especially in remote areas and complex terrains. • Prerequisites: High-quality visual feature extraction and a reliable UAV platform are required. • Industry Impact: By enhancing inspection efficiency and reducing operational risks, the framework is poised for widespread adoption in the energy transportation sector, ensuring pipeline structural integrity and safety.

Despite its excellent performance in experiments, the framework's performance may be affected in extreme weather conditions, such as strong winds or heavy rain. Additionally, the method relies on high-quality visual feature extraction, so performance may degrade under poor lighting conditions or low image quality. Future research directions include further optimizing the ESKF-PRE-VMPC framework to enhance its robustness and adaptability in extreme environmental conditions.