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

## TL;DR

This paper introduces a robot controller using a reinforcement learning algorithm to maintain constant force during grinding, improving adaptability without needing detailed environmental models.

## Contribution

The novel contribution is a robot constant force grinding controller based on proximal policy optimization, which reduces reliance on environmental models.

## Key findings

- The controller uses proximal policy optimization to train a model for grinding force compensation.
- Simulation results show the controller achieves constant force grinding without prior environmental modeling.
- The method demonstrates environmental adaptability in robot grinding tasks.

## Abstract

In order to solve the problems of high dependence on the accuracy of environmental model and poor environmental adaptability of traditional control methods, the robot constant force grinding controller that based on proximal policy optimization was proposed. Training the controller model between grinding force difference and end-effector compensation displacement using the proximal policy optimization algorithm. Complete compensation using robot inverse kinematics. In order to validate the algorithm, a simulation model of the grinding robot with perceivable force information is established. The simulation results demonstrate that the controller trained using this algorithm can achieve constant force grinding without setting up the environment model in advance and has some environmental adaptability.