Towards Task Planning with Learned Manipulation Skills

This colloquium talk explores the integration of task planning with learned manipulation skills in robotics. Join Sebastian Castro from the Robotics and AI Institute as he examines how the growing repository of fundamental robotics techniques can be composed to achieve high-level goals. Discover the landscape of tools enabling task-level planning and the behavior abstractions needed to ground these plans into executable skills in real-world environments. Castro discusses promising research directions for composing learned manipulation skills while reducing dependency on domain-specific, hand-engineered solutions. With experience spanning Cornell University, MathWorks, MIT CSAIL, Boston Dynamics, and PickNik Robotics, Castro brings valuable insights to this 52-minute presentation from the 2025 Winter Robotics Colloquium held on February 21, 2025. The video includes closed captioning for accessibility.