This article presents a general state of the art of the Transfer Learning (TL) and Reinforcement Learning (RL) intersection with a special focus on applications on mobile robotics. First, we study the key ideas in TL—transfer of feature representation, transfer of model, and transfer of policy—and explore how these ideas mitigate the challenges posed by long training times, and high sample complexity associated with RL. Indeed, a citation analysis conducted between 2015 and 2024 confirms increasing interest in this integrative approach in academia. Expanding on this trend, we introduce and contrast different TL techniques applied to the core mobile robot tasks of navigation, obstacle avoidance, and target tracking. Next, we discuss existing simulation tools and real-world datasets, including NVIDIA Isaac Sim, PyBullet, and RoboNet, that are commonly used as performance benchmarks for TL-based RL methods. For each area, important evaluation metrics used are delineated, including reduction in sample complexity, robustness to domain shifts, etc., to reflect the efficacy and generalizability of TL strategies. Finally, we discuss present-day obstacles, such as excessive dependence upon simulation and requirements for more sophisticated multi-robot benchmarks and we highlight future avenues for extending and refining TL techniques for mobile robotics.