Variable-Parameter Impedance Control of Manipulator Based on RBFNN and Gradient Descent

Area of Science:

• Robotics
• Control Systems Engineering
• Artificial Intelligence

Background:

• Accurate force and position control are crucial for manipulator grasping tasks to prevent object damage.
• Existing hybrid force/position control methods are computationally intensive.
• Adaptive control is needed for manipulators interacting with objects of varying stiffness.

Purpose of the Study:

• To propose a computationally efficient variable-parameter impedance control method for manipulators.
• To achieve concurrent and precise force and position control during grasping.
• To reduce the need for manual impedance parameter tuning.

Main Methods:

• A position-based impedance control structure integrating iterative learning control and gradient descent.
• Design of a sliding mode controller to handle system uncertainties and perturbations.
• Utilizing a Radial Basis Function Neural Network (RBFNN) for system identification.
• Iterative adjustment of impedance parameters via a gradient descent method.

Main Results:

• The proposed method achieves precise force and position control in manipulator grasping tasks.
• Demonstrated efficacy through simulations and experimental validation.
• Successfully adapted to objects with varying stiffness without manual parameter adjustments.

Conclusions:

• The variable-parameter impedance control method offers a superior alternative to traditional methods.
• The integration of RBFNN and gradient descent significantly improves control accuracy and adaptability.
• This approach enhances the safety and efficiency of robotic manipulation tasks.