Image-Based Visual Servoing of Manipulators With Unknown Depth: A Recurrent Neural Network Approach

Area of Science:

• Robotics
• Computer Vision
• Control Systems

Background:

• Image-based visual servoing (IBVS) is crucial for intelligent robotic manipulation using visual feedback.
• Traditional IBVS methods often require depth information, complicating implementation.
• The interaction matrix in conventional IBVS necessitates knowledge of depth parameters.

Purpose of the Study:

• To propose a novel IBVS method for manipulators that does not require depth estimation.
• To simplify the visual servoing process by removing the dependency on depth information.
• To develop a robust control law for visual servoing without explicit depth calculation.

Main Methods:

• A novel transformation is employed to convert the IBVS problem into a convex optimization problem.
• The method utilizes the properties of the associated image Jacobian.
• A recurrent neural network with global asymptotic convergence is developed to solve the optimization problem.

Main Results:

• The proposed method successfully achieves IBVS for manipulators without requiring depth estimation.
• A dynamic neural control law is derived, demonstrating efficacy through theoretical guarantees and simulations.
• The approach handles kinematic, joint, and other depth-independent constraints effectively.

Conclusions:

• The novel IBVS method offers a simplified and effective approach to robotic manipulation.
• Eliminating depth estimation enhances the practicality and applicability of visual servoing.
• The developed neural network and control law provide a robust solution for depth-free IBVS.