Retail environments often create stress on shoppers due to confusing layouts, difficulty to locate products and crowding. These stresses can prevent shoppers from purchasing all their intended items. This can be detrimental to their well being as well as financially damaging to the store. To address these concerns, this thesis presents a retail robot architecture that can explore and locate requested products in an unknown, crowded, environment. Furthermore, a novel mapping algorithm entitled contextSLAM is presented, which creates contextually rich maps of crowded environments to aid in localisation and exploration. ContextSLAM was validated in simulation then implemented in the full retail architecture. The full system was then validated through a battery of experiments where it was tasked with finding requested products in both a mock retail environment and a grocery store. These experiments were conducted in both uncrowded and crowded settings to ensure repeatability and robustness.