In recent years, autonomous robots have been increasingly investigated for operation in complex real-world environments, such as search and rescue, inspection, and exploration of areas that are difficult to access or reproduce. In these scenarios, traditional wheeled robots are often limited by terrain conditions, while legged robots provide greater flexibility due to their ability to interact with the environment through discrete contact points. This thesis focuses on the development of a locomotion system for a real reconfigurable hexapod robot using Reinforcement Learning. Hexapod robots represent a suitable platform for this objective because their multi-legged structure provides stability, redundancy, and the possibility of studying locomotion under different morphological configurations.