A controller of robot constant force grinding based on proximal policy optimization algorithm.

Area of Science:

• Robotics
• Control Engineering
• Machine Learning

Background:

• Traditional robot grinding control methods often rely heavily on accurate environmental models.
• Poor environmental adaptability limits the performance of existing control strategies in dynamic grinding scenarios.

Purpose of the Study:

• To develop a robot constant force grinding controller that overcomes the limitations of traditional methods.
• To enhance environmental adaptability and reduce dependence on precise environmental models in robot grinding operations.

Main Methods:

• Proposed a robot constant force grinding controller based on proximal policy optimization (PPO).
• Trained the controller model using the PPO algorithm, mapping grinding force difference to end-effector compensation displacement.
• Utilized robot inverse kinematics for motion compensation.

Main Results:

• A simulation model of a grinding robot with force perception was established for validation.
• The trained controller successfully achieved constant force grinding without prior environmental modeling.
• Demonstrated notable environmental adaptability of the proposed control algorithm.

Conclusions:

• The proximal policy optimization-based controller offers a robust solution for constant force grinding.
• This approach significantly reduces the need for environmental models, enhancing practical applicability.
• The controller exhibits promising environmental adaptability for real-world grinding tasks.