Learning to Compose Hierarchical Object-Centric Controllers for Robotic Manipulation

Manipulation tasks can often be decomposed into multiple subtasks performed\nin parallel, e.g., sliding an object to a goal pose while maintaining contact\nwith a table. Individual subtasks can be achieved by task-axis controllers\ndefined relative to the objects being manipulated, and a set of object-centric\ncontrollers can be combined in an hierarchy. In prior works, such combinations\nare defined manually or learned from demonstrations. By contrast, we propose\nusing reinforcement learning to dynamically compose hierarchical object-centric\ncontrollers for manipulation tasks. Experiments in both simulation and real\nworld show how the proposed approach leads to improved sample efficiency,\nzero-shot generalization to novel test environments, and simulation-to-reality\ntransfer without fine-tuning.\n