Machine Learning and Friends Lunch: David Held, Relational Learning for Robot Manipulation
Content
Speaker
David Held (Carnegie Mellon University)
Abstract
Robots in factories today are typically confined to interact with rigid objects with known object models. How can we bring robots into the more diverse, unstructured settings of our daily lives, where objects are often deformable, articulated, or varied in shape and appearance, while maintaining high levels of performance?
I argue that advancing robot capabilities requires learning methods that reason about relationships: (1) relationships between the gripper and the scene, and (2) relationships between objects being manipulated and other objects in the environment. I will show that such relational reasoning enables robots to perform complex tasks, such as deformable and articulated object manipulation, precise insertion, and non-prehensile manipulation, while generalizing to unseen objects and unseen configurations. By incorporating relational reasoning, we can achieve robust performance on challenging robot manipulation tasks.
Speaker Bio
David Held is an Associate Professor at Carnegie Mellon University in the Robotics Institute and is the director of the RPAD lab: Robots Perceiving And Doing. His research focuses on perceptual robot learning, i.e. developing new methods at the intersection of robot perception and planning to teach robots how to manipulate novel, perceptually challenging, and deformable objects. Prior to coming to CMU, David was a post-doctoral researcher at U.C. Berkeley, and he completed his Ph.D. in Computer Science at Stanford University. David also has a B.S. and M.S. in Mechanical Engineering at MIT. David is a recipient of the Google Faculty Research Award in 2017 and the NSF CAREER Award in 2021. David is also leading a MURI team on the topic of “Cognitive and Neuroscience‐Inspired Problem‐Solving for Autonomous Systems in Physical Environments.”