Project Page

PO-PDDL: Learning Symbolic POMDPs from Visual Demonstrations for Robot Planning Under Uncertainty

PO-PDDL learning pipeline
We introduce PO-PDDL, a symbolic formulation of POMDPs, and propose a pipeline for learning PO-PDDL models from visual demonstrations. The resulting PO-PDDL domains enable online-belief space planning under uncertainty.

Video Presentation

Visual Demonstrations

close the yellow drawer

place the cup with water on the top of the green drawer

pick up the red block and place the red block on the top of the yellow drawer

open the yellow drawer

place the pink cup on the top of the green drawer

click to view more demonstrations

Generated Domain from Demonstrations

Online Planning with the Learned PO-PDDL Domain