The Quest for Robot Intelligence

The quest for robot intelligence has driven decades of research, as we strive to create robots that can do our household chores, and learn new tasks flexibly and autonomously even from non-expert users. Early approaches relied on teleoperation and manual programming, but over time, the focus has shifted to enabling robots to learn by themselves, through interaction, combining strategies like imitation and reinforcement learning. While imitation allows robots to replicate human actions, reinforcement learning enables them to improve on their own, through trial and error, often done in simulation first, before transferring the learned skills to the real world. In this talk, we will trace the evolution of these learning paradigms, highlighting key milestones in robot intelligence. We will explore practical examples such as object manipulation, bipedal locomotion, and VR/AR human-robot interaction, and examine the strengths and limitations of current learning algorithms. As we look to the near future, we are excited by the potential to create robots that transcend imitation, unlocking true autonomy in learning and decision-making.