Federated Learning (FL), is a machine learning (ML) technique that enables a decentralized method for multiple users to share training results to collaboratively learn, while keeping secure personal data. With the increasing amount of devices that are able to collect data and train ML models, FL has gain interest as it is capable to overcome challenges of traditional methods like security and scalability. In this paper, we propose a federated framework based on ROS 2, a popular robotic framework used to communicate the different elements of a robotic system.