Model-Free Reinforcement Learning for Parabolic Trajectory Optimization in Robotic Arms

Abstract

Robotic arms are widely employed in applications that require smooth motion and energy-efficient operation, particularly in tasks such as object throwing and liquid dispensing, where movement often follows a curved path toward a target point. However, conventional trajectory planning methods that rely on predefined mathematical equations may not accurately represent real-world robotic systems due to uncertainties and payload variations. This study aims to optimize the trajectory of a robotic arm moving along a parabolic path using reinforcement learning and to evaluate whether this approach can successfully learn improved trajectory patterns during motion. The research integrates initial classical physics principles for curved motion with a reinforcement learning framework to enhance trajectory following toward a desired point. The findings indicate that reinforcement learning can effectively learn optimized trajectory paths and improve the motion performance of the robotic arm. The study concludes that reinforcement learning offers a promising approach for achieving smoother robotic motion with satisfactory energy efficiency under dynamic conditions. This work contributes to the advancement of intelligent motion planning by demonstrating the potential of reinforcement learning to improve trajectory optimization in robotic systems operating under practical uncertainties.