[RPAD Website][Publications][Lab Members]

To maintain fairness, transparency, and academic integrity, I do not accept gifts from students, collaborators, vendors, or other external parties. This includes monetary gifts, gift cards, and other material items. Even well-intentioned gestures can create the appearance of bias or a conflict of interest, particularly in academic and professional settings. If you wish to express appreciation, a written note or email is always appreciated.

David Held is an Associate Professor in the Robotics Institute at Carnegie Mellon University and the director of the Robots Perceiving and Doing (RPAD) Lab. His research focuses on robot learning for physical reasoning and manipulation. He develops methods at the intersection of perception and planning that enable robots to manipulate novel and deformable objects and recover from failures in complex, unstructured environments. Prior to joining CMU, David was a postdoctoral researcher at UC Berkeley. He received his Ph.D. in Computer Science from Stanford University and his B.S. and M.S. degrees in Mechanical Engineering from MIT. He leads a multidisciplinary DoD MURI project on enabling robots to overcome unexpected challenges and is a recipient of the NSF CAREER Award and the Google Faculty Research Award.

You can also download my CV (last updated Aug 17, 2026).

My research lies at the intersection of robotics, machine learning, and computer vision.

I am interested in developing new methods for robotic perception and control that can allow robots to operate in the complex environments of our daily lives. I have applied the idea of perceptual robot learning to improve a robot's capabilities in two domains: object manipulation and autonomous driving. In the realm of object manipulation, I am developing methods for robots to learn to manipulate novel objects, perceptually challenging objects (e.g. transparent and specular), and deformable objects (e.g. cloth). Regarding autonomous driving, I am developing methods for self-supervised learning and semi-supervised learning (e.g. learning from unlabeled data). Solving these challenges requires rethinking robot perception and control algorithms to handle these types of tasks.