Obstacle Avoidance Control Method for Robotic Assembly Process Based on Lagrange PPO

The actual 3C assembly of robotic arms in factories often involves high-risk operations that need to be secured to prevent accidents. However, traditional reinforcement learning (RL) methods tend to pursue high rewards while ignoring behaviors that may lead to safety risks. To ensure the safety and efficiency of the industrial robot production process, we use the Lagrange Proximal Policy Optimization (L-PPO) algorithm to control the robotic arm. This method incorporates constraints into the agent learning and decision-making process, and the robotic arm considers obstacle locations and characteristics during task execution to ensure that its behavior is within safe limits. First, sensors are used to obtain information about the current state of the robotic arm and the obstacles in the environment. Then, L-PPO is used to train the robot arm’s policy network. This network maps the current state to the robot arm’s action space to generate the robot arm’s action strategy. By interacting with the environment and collecting trajectory data, the strategy of the robotic arm is continuously improved by performing gradient updates using the L-PPO. Experimental results show that by using the L-PPO, the robotic arm is able to efficiently avoid obstacles and successfully reach the target. The method showed better performance on the obstacle avoidance task compared to traditional RL methods.