Quadruped Robot Prototype Based On Artificial Intelligence Using Deep Reinforcement Learning Through The ROS Framework

Soto J.; La Rosa Lama G.; Arcaya J.I. · Proceedings - IEEE CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies, ChileCon (2025) · DOI: 10.1109/chilecon66915.2025.11476093

Resumen

This paper presents a quadruped robot prototype controlled through deep reinforcement learning, implemented using the Robot Operating System (ROS). The system is structured into modular nodes for sensor acquisition, control, and actuation. Control models based on Q-Table and Deep Q-Network (DQN) were developed and trained in Gym environments for tasks such as posture adjustment and translation. Experimental results demonstrate that DQN offers better adaptability and performance in complex scenarios. The integration of ROS, reinforcement learning, and visual perception provides a scalable solution for autonomous robots in unstructured environments.

Instituciones: Universidad de Piura