Inverted Pendulum Control with a Robotic Arm using Deep Reinforcement Learning

摘要

Inverted pendulum control is a benchmark control problem that researchers have used to test the new control strategies over the past 50 years. Deep Reinforcement Learning Algorithm is used recently on the inverted pendulum on a straightforward form. The inverted pendulum had only one degree of freedom and was moving on a plane. This paper demonstrates a successful implementation of a deep reinforcement learning algorithm on an inverted pendulum that rotates freely on a spherical joint with an industrial 6 degrees freedom robot arm. This research used the Deep Reinforcement Learning algorithm in Robot Operating System (ROS) and Gazebo Simulation. Experimental results show that the proposed method achieved promising outputs and reaches the control objectives. We were able to control the inverted pendulum upward for 30 and 20 seconds in two case studies. Two other significant novelties in this research are using an inertial measurement unit (IMU) on the tip of the pendulum, that will facilitate implementation on the real robot for future work and different reward functions in comparing to past publications that enable continuous learning and mastering control in a vertical position