Hierarchical Robot Planning with Black-Box Skills
A novel method for task and motion planning (TAMP) merges various existing robot skills—such as learned, force-controlled, and black-box policies—within a hierarchical planning framework. This technique, known as Task and Skill Planning (TASP), utilizes Composable Interaction Primitives (CIPs) to create motion plans that connect successive skills, allowing for refinement during planning and adjustments during execution. TASP maintains the object-centric failure reasoning characteristic of TAMP solvers. Real-world experiments were conducted to validate the approach using both a bimanual manipulator and a mobile manipulator. This research is documented in the arXiv preprint 2504.17901.
Key facts
- Task and Skill Planning (TASP) integrates heterogeneous robot skills into a hierarchical planner.
- Skills include learned, force-controlled, and black-box policies.
- Composable Interaction Primitives (CIPs) synthesize motion plans bridging skills.
- TASP preserves object-centric failure reasoning.
- Validated on bimanual and mobile manipulators in real-world experiments.
- Published as arXiv:2504.17901.
Entities
Institutions
- arXiv