caelan/pddlstream

Extends PDDL with streams—declarative sampling procedure specifications—enabling domain-independent planning that treats samplers as blackboxes without requiring their implementation details. Implements multiple algorithms (Incremental, Focused, Adaptive) that interleave symbolic planning with constraint satisfaction, with the Adaptive approach optimizing for domains with multiple sampling pathways like robotic manipulation. Integrates with Fast Downward for classical planning and includes PyBullet-based examples for task-and-motion planning (TAMP) on robot platforms including PR2 and Kuka arms.

468 stars. No commits in the last 6 months.