Dynamic Ball Balancing Using Deep Deterministic Policy Gradient (DDPG)-Controlled Robotic Arm for Precision Automation

This paper presents a reinforcement learning (RL)-based solution for dynamic ball balancing using a robotic arm controlled by the Deep Deterministic Policy Gradient (DDPG) algorithm. The problem addressed is maintaining ball stability under external disturbances in automated manufacturing. The proposed solution enables adaptive, precise control on flat surfaces. The research contribution is a comparative evaluation of DDPG and Soft Actor-Critic (SAC) algorithms for trajectory control and stabilization. A simulated environment is used to train the RL agents across multiple initial ball positions. Key performance metrics-settling time, rise time, overshoot, and steady-state error-are analyzed. Results show DDPG outperforms SAC with smoother trajectories, ~25% faster settling times, and significantly lower overshoot and steady-state errors. Visual analysis confirms that DDPG consistently drives the ball to the center with minimal deviation. These findings highlight DDPG’s advantages in control accuracy and stability. In conclusion, the DDPG-based approach proves highly effective for precision automation tasks where fast, stable, and reliable control is essential.