Time: 7:00-9:00 p.m., Wednesday, March 4th
Location: Armenise Amphitheater at Harvard Medical School, 200 Longwood Avenue, Boston (link to directions)
Speakers: Maria Bauza Villalonga and Ferran Alet
Graphics: Rebecca Senft
Robotic hardware has made enormous progress recently, with drones being commonplace and Boston dynamics releasing impressive videos. However, robots are only present in very limited environments, which only require very repetitive actions. Through the lens of the Amazon Robotics Challenge, which required building an autonomous system that could do pick and place in a warehouse, we will discuss what robots can and cannot currently do. We will also review how progress in deep learning affects robotics through the biggest recent advancements: reliable vision for self-driving cars, AlphaZero beating the world champion at Go, and OpenAI successfully transferring a robotic policy learned in simulation to the real world. Finally, so far progress in Machine Learning has mostly involved computer vision and large datasets which limits its applicability. We will introduce new research directions that go beyond vision, by teaching robots the sense of touch and generalize with few amounts of data by doing meta-learning.