Abstract:
Robots need fine sensorimotor skills to pick up any object, use tools, assemble parts, and recover from mistakes—but collecting enough robot data to train such skills is painfully slow and expensive. Meanwhile, the internet overflows with videos of humans performing exactly these tasks. In this talk, I will present our lab’s work on learning robot manipulation skills directly from in-the-wild human videos, bypassing the need for massive robot-only datasets. By leveraging diverse, unstructured demonstrations, our methods generalize to novel objects and scenarios beyond the small amount of robot training data used for fine-tuning. I will also outline key open challenges toward truly generalist robots that we are working on in my lab, from better policy architectures and
multi-sensory integration to lifelong learning and hardware design.
Biography:
Jeannette Bohg is an Assistant Professor of Computer Science at Stanford University. She was a group leader at the Autonomous Motion Department (AMD) of the MPI for Intelligent Systems until September 2017. Before joining AMD in January 2012, Jeannette Bohg was a PhD student at the Division of Robotics, Perception and Learning (RPL) at KTH in Stockholm. In her thesis, she proposed novel methods towards multi-modal scene understanding for robotic grasping. She also studied at Chalmers in Gothenburg and at the Technical University in Dresden where she received her Master in Art and Technology and her Diploma in Computer Science, respectively. Her research is at the intersection of Robotics, Machine Learning and Computer Vision applied to problems in Autonomous Robotic Manipulation. Her lab seeks to understand the underlying principles of robust sensorimotor coordination by implementing them on robots.
Jeannette Bohg has received several Early Career and Best Paper awards, most notably the 2019 IEEE Robotics and Automation Society Early Career Award, the 2020 Robotics: Science and Systems Early Career Award and the 2023 Sloan Research Fellowship.