TL;DR
This paper introduces a novel uncalibrated visual servoing system for underwater robots with attached manipulators. It tackles the challenge of unknown camera parameters, robot kinematics, and external disturbances by employing a linear separation approach for uncertainties, an adaptive controller with a disturbance observer, and the gradient projection method to optimize robot movements. The proposed system demonstrates accurate and smooth trajectory tracking in simulations, even under challenging underwater conditions.
What Problem Does This Paper Address?
Underwater vehicles (UVMSs) equipped with manipulators are crucial for oceanic exploration and intervention tasks. To perform these tasks autonomously, they rely on visual servoing, where cameras guide the manipulator's actions. However, achieving precise visual servoing is difficult due to several factors:
- Uncalibrated Cameras: Underwater environments make camera calibration complex and costly. Traditional visual servoing methods often require precise camera intrinsic and extrinsic parameters, which are hard to obtain reliably.
- Unknown UVMS Kinematics: The precise kinematic relationships of the underwater vehicle and its manipulator can also be uncertain.
- External Disturbances: Underwater currents and other environmental factors introduce time-varying disturbances that can significantly affect the robot's performance.
- Complex Dynamics: UVMSs are free-floating multibody systems, meaning the manipulator's movements strongly influence the vehicle's dynamics, and vice-versa. This coupling is difficult to model precisely.
Existing methods often struggle to address all these uncertainties simultaneously, especially in dynamic-based visual servoing for UVMSs. This paper aims to bridge this gap by proposing a robust uncalibrated visual servoing scheme that can handle these challenges.
Methods
The authors propose a multi-faceted approach to tackle the uncalibrated visual servoing problem:
- Linear Separation for Uncertainties: A novel linear separation method is introduced to group various uncertainties (vision sensor parameters, UVMS kinematics, feature position) into manageable constant vectors. This approach is designed to be applicable to other free-floating articulated manipulators as well.
- Adaptive Controller with Gradient Projection Method (GPM): A nonlinear adaptive controller is developed to drive image errors to zero. This controller estimates the unknown uncertainty vectors. Crucially, the Gradient Projection Method (GPM) is integrated to optimize the robot's "restoring moments" by exploiting the manipulator's kinematic redundancy. This helps in managing the complex dynamics and reducing unwanted movements.
- High-Order Disturbance Observer (HODO): To counter time-varying external disturbances (like currents), a high-order disturbance observer is employed. This observer estimates and compensates for these unknown disturbances, allowing for more stable control.
- Composite Controller: A composite controller is designed, integrating the adaptive laws, the GPM-optimized reference velocity, and the disturbance compensation from the HODO.
- Lyapunov Stability Analysis: The theoretical stability of the closed-loop system is rigorously proven using Lyapunov theory, ensuring that the image errors converge to zero.
Datasets
Not discussed in the paper. The evaluation is performed using numerical simulations.
Key Insights and Ideas
- Uncertainty Grouping: The linear separation approach is a clever way to manage multiple sources of uncertainty by consolidating them into estimable constant vectors, simplifying the control problem.
- GPM for Restoring Moments: Using the Gradient Projection Method to optimize restoring moments is a key innovation for UVMSs. It allows the robot to actively manage its dynamics and improve tracking performance, especially in the presence of complex coupling between the vehicle and manipulator.
- HODO for Robustness: The adoption of a High-Order Disturbance Observer is critical for real-world underwater applications, where constant disturbances are a given. Its ability to handle time-varying disturbances without assuming them to be constant is a significant advantage.
- Unified Approach: The paper successfully integrates uncalibrated vision, adaptive control, disturbance rejection, and dynamic optimization into a single, coherent framework for UVMS visual servoing.
Key Results
The proposed scheme was evaluated through numerical simulations on a 9-DOF UVMS with an eye-in-hand camera. The system was tasked with tracking a circular trajectory on the image plane while subjected to time-varying disturbances.
- Accurate and Smooth Tracking: The simulations demonstrated that the proposed scheme achieves accurate and smooth tracking of the desired circle trajectory.
- Convergence of Errors: The image space tracking errors (both position and velocity) converged to near zero.
- Disturbance Rejection: The HODO effectively estimated and compensated for external disturbances, as shown by the bounded disturbance estimation errors.
- GPM Benefits: The GPM-based reference velocity term showed a significant reduction in the influence of restoring moments compared to a method without GPM.
- Performance Comparison: The proposed method (with HODO and GPM) outperformed a baseline without the HODO in terms of trajectory tracking accuracy and smoothness.
Limitations and Future Work
- Simulation-Based: The effectiveness of the proposed method is demonstrated through numerical simulations. Real-world experimental validation on an actual UVMS is needed to confirm its performance in practice.
- Assumption of Non-Singular Jacobian: The method assumes that the depth-Jacobian matrix is non-singular. While this is a common assumption in uncalibrated visual servoing, strategies to handle potential singularities might be necessary for broader applicability.
- Parameter Estimation: While the method estimates unknown parameters, the convergence speed and accuracy of these estimations in highly dynamic or noisy real-world scenarios might require further investigation.
The authors suggest that the proposed linear separation method can be applied to other free-floating articulated manipulators. Future work could explore extending this framework to more complex underwater tasks and environments.
Summary updated