This project aims to tackle a critical challenge in robotics: helping mobile robots to autonomously learn robust navigation strategies from unstructured, real-world data in human-populated environments. Building on the success of the “Lindsey: the tour guide robot” project (2018–2023), which involved real-world deployments in public spaces, we will introduce groundbreaking methodologies for learning from suboptimal demonstrations and in-situ data generated through natural human-robot interactions. Unlike conventional methods that rely on synthetic or highly curated datasets, our approach leverages the unpredictability of real-world settings and untrained human assistance to develop adaptive and scalable navigation policies.