Simulation and Control of a Robotic Arm to Intercept a Ping-Pong Ball Using Deep Reinforcement Learning and Computer Vision

Hernandez S.R.; Ramos N.M. · 2026 12th International Conference on Mechatronics and Robotics Engineering, ICMRE 2026 (2026) · DOI: 10.1109/icmre69538.2026.11533984

Resumen

This article presents the simulation and control of a five-degree-of-freedom robotic arm designed to intercept a ping pong ball in real time. The system incorporates servo motors and a BLDC motor that provide the required torque and speed to perform dynamic and precise movements. For ball detection and tracking, a computer vision system based on 120 FPS cameras is employed, capable of estimating the ball's trajectory and generating a target interception point for the arm. Initially, a position control system will be implemented to enable the arm to follow predefined trajectories based on the predicted ball path. Subsequently, a deep reinforcement learning model will be integrated and trained in simulation to avoid physical environment errors. The system's performance will be validated by comparing its behavior in simulation and realworld conditions, focusing on responsiveness to fast and nonlinear trajectories.

Instituciones: Universidad San Ignacio de Loyola