Modularity for Robot Manipulation

Title: Modularity for Robot Manipulation
Abstract:
In the future, we aim to build robots with the robustness and versatility needed to operate in open-world environments such as homes, service industries, and hospitals. Achieving this vision requires robots to learn manipulation skills that generalize across diverse objects and task scenarios. Modularity is a key principle for enabling such adaptability. In this talk, I will present recent work from my lab on leveraging modularity to structure manipulation tasks for more efficient learning. I will discuss how robots can acquire modular skill representations tailored to specific tasks, enabling rapid adaptation. By exploiting intra-skill modularity, robots can efficiently transfer behaviors between different objects using foundation models. I will subsequently describe our work on learning the scope of skills and planners. By understanding the scope of their abilities, robots can identify their own limitations and address them in a systematic manner. Finally, I will highlight ongoing efforts to develop modular robot hardware, with a focus on new designs for robotic hands.
Bio:
Dr. Oliver Kroemer is an Associate Professor in the Robotics Institute at Carnegie Mellon University (CMU), where he leads the Intelligent Autonomous Manipulation Lab. His research focuses on developing algorithms and representations that enable robots to acquire versatile and robust manipulation skills. Before joining CMU, Dr. Kroemer was a postdoctoral researcher at the University of Southern California (USC). He received his Bachelor’s and Master’s degrees in Engineering from the University of Cambridge in 2008. From 2009 to 2011, he conducted his doctoral research at the Max Planck Institute for Intelligent Systems and completed his Ph.D. in 2014 at the Technische Universität Darmstadt with a dissertation on machine learning for robot grasping and manipulation.