Abstract: Given the advancements in artificial intelligence over the last decade along with the significant decrease in hardware cost, indoor service robots such as Roombas are increasingly becoming a part of our lives.  In this talk, I will highlight my lab's recent efforts towards developing AI for such autonomous agents. First, I will discuss an accurate data-driven model for simulating human interactions in a crowd rooted in statistical mechanics. Second, I will show how such a model can be extended to develop an optimization-based method for fast, anticipatory collision avoidance applicable to a wide range of mobile robots. Third, I will discuss how we can enhance the high-level decision making capabilities of such robots, including introducing novel VAE-based approaches for accurate agent trajectory prediction and shaping human-inspired navigation decisions via knowledge distillation and online learning. I will conclude the talk with a recent real-world study that we ran to gain more insights about how humans interact and behave in the presence of small service indoor robots, and discuss limitations of existing robot controllers and potential avenues for future work.