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

2026-06-14

AI Summary

Real-world robotic task planning must account for action stochasticity and partial observability, yet constructing practical POMDP models remains challenging. This work proposes PO-PDDL—a symbolic POMDP representation that preserves the relational structure of PDDL while being compatible with large language models—and presents the first extension of PDDL to support belief states and partial observability. By automatically learning state trajectories from robot execution videos and detecting observational inconsistencies, the system jointly performs visual state reconstruction, symbolic trajectory inference, and learning of stochastic transition and observation models, enabling online belief-space planning. Experiments demonstrate that the approach significantly outperforms existing PDDL- and POMDP-based learning methods on real long-horizon manipulation tasks, achieving robust planning under uncertainty at substantially lower computational cost.

Abstract

Real-world robot task planning must operate under both stochastic action execution and partial observability, yet constructing Partially Observable Markov Decision Process (POMDP) models for real robotics domains remains difficult and labor-intensive. We introduce PO-PDDL, a symbolic formulation of POMDPs that preserves the relational structure and LLM-friendly syntax of the Planning Domain Definition Language (PDDL), while explicitly modeling partial observability, stochasticity, and beliefs. Building on this formulation, we propose a demonstration-driven pipeline for learning PO-PDDL models. The proposed method reconstructs latent symbolic state trajectories from real-robot execution videos, identifies partial observability via inconsistencies between inferred states and visual observations, and learns stochastic transition and observation models accordingly. The resulting PO-PDDL domains are reusable across tasks and enable online belief-space planning under both perception and execution uncertainty. Experiments on real-world long-horizon manipulation tasks show that our method consistently outperforms existing PDDL and POMDP model-learning approaches, achieving robust task planning under uncertainty with significantly lower planning cost.

Problem

• POMDP
• partial observability
• robot planning
• uncertainty
• symbolic modeling

Innovation

• PO-PDDL
• symbolic POMDPs
• visual demonstrations
• belief-space planning
• partial observability