Model predictive control for constrained robot manipulator visual servoing tuned by reinforcement learning.

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Constrained image-based visual servoing (IBVS) presents challenges in robot manipulator control.
• Model predictive control (MPC) offers a framework for handling system constraints in control tasks.

Purpose of the Study:

• To propose a novel model predictive control (MPC) strategy tuned by reinforcement learning (RL) for constrained image-based visual servoing (IBVS) of robot manipulators.
• To enhance the speed and stability of robot manipulator responses in visual servoing tasks.

Main Methods:

• Model predictive control (MPC) is employed to formulate the IBVS task as a nonlinear optimization problem under system constraints.
• A depth-independent visual servo model is utilized as the predictive model within the MPC framework.
• A deep-deterministic-policy-gradient (DDPG) based reinforcement learning (RL) algorithm is used to train the MPC objective function weight matrix.

Main Results:

• The proposed RL-tuned MPC strategy generates sequential joint signals for robot manipulators.
• The controller enables the robot manipulator to achieve desired states more rapidly.
• Simulation experiments demonstrate the efficacy and stability of the developed control strategy.

Conclusions:

• The integration of RL with MPC provides an effective approach for constrained IBVS.
• The proposed method offers improved performance in terms of speed and stability for robot manipulator control.
• This strategy holds promise for advanced robotic applications requiring precise visual servoing.