Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning

Caelan Reed Garrett, Tomás Lozano-Pérez, Leslie Pack Kaelbling

TL;DR

PDDLStream addresses planning in continuous spaces with complex constraints by coupling symbolic PDDL planning with black-box samplers called streams. It formalizes streams, their inputs/outputs and certified facts, and reduces planning to a sequence of finite PDDL problems. Four algorithms are introduced, with Adaptive balancing exploration of new plans against sampling of existing ones, significantly improving performance on tightly constrained and cost-sensitive problems. The framework is validated in robotic manipulation and multi-robot rover domains and demonstrated on real PR2 tasks, highlighting practical applicability to real-world robotics.

Abstract

Many planning applications involve complex relationships defined on high-dimensional, continuous variables. For example, robotic manipulation requires planning with kinematic, collision, visibility, and motion constraints involving robot configurations, object poses, and robot trajectories. These constraints typically require specialized procedures to sample satisfying values. We extend PDDL to support a generic, declarative specification for these procedures that treats their implementation as black boxes. We provide domain-independent algorithms that reduce PDDLStream problems to a sequence of finite PDDL problems. We also introduce an algorithm that dynamically balances exploring new candidate plans and exploiting existing ones. This enables the algorithm to greedily search the space of parameter bindings to more quickly solve tightly-constrained problems as well as locally optimize to produce low-cost solutions. We evaluate our algorithms on three simulated robotic planning domains as well as several real-world robotic tasks.

PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning

TL;DR

PDDLStream addresses planning in continuous spaces with complex constraints by coupling symbolic PDDL planning with black-box samplers called streams. It formalizes streams, their inputs/outputs and certified facts, and reduces planning to a sequence of finite PDDL problems. Four algorithms are introduced, with Adaptive balancing exploration of new plans against sampling of existing ones, significantly improving performance on tightly constrained and cost-sensitive problems. The framework is validated in robotic manipulation and multi-robot rover domains and demonstrated on real PR2 tasks, highlighting practical applicability to real-world robotics.

Abstract

Many planning applications involve complex relationships defined on high-dimensional, continuous variables. For example, robotic manipulation requires planning with kinematic, collision, visibility, and motion constraints involving robot configurations, object poses, and robot trajectories. These constraints typically require specialized procedures to sample satisfying values. We extend PDDL to support a generic, declarative specification for these procedures that treats their implementation as black boxes. We provide domain-independent algorithms that reduce PDDLStream problems to a sequence of finite PDDL problems. We also introduce an algorithm that dynamically balances exploring new candidate plans and exploiting existing ones. This enables the algorithm to greedily search the space of parameter bindings to more quickly solve tightly-constrained problems as well as locally optimize to produce low-cost solutions. We evaluate our algorithms on three simulated robotic planning domains as well as several real-world robotic tasks.

Paper Structure

This paper contains 28 sections, 3 theorems, 9 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. PDDLStream
  4. Streams
  5. Domain Description
  6. Example Domains
  7. Rovers Domain
  8. PDDLStream Algorithms
  9. Incremental Algorithm
  10. Optimistic Algorithms
  11. Focused Algorithm
  12. Binding and Adaptive Algorithms
  13. Binding Algorithm
  14. Example Execution
  15. Adaptive Algorithm
  16. ...and 13 more sections

Key Result

Theorem 1

PDDLStream plan existence is undecidable.

Figures (4)

  • Figure 1: Left: Domain 1 (with 5 blocks). Right: A real-world robot planning to "serve a meal" on the brown tray.
  • Figure 2: Domain 3 (with 4 objectives).
  • Figure 3: From left to right: Domain 1 success percent, Domain 1 mean runtime, and Domain 2.
  • Figure 4: From left to right: Domain 3 success percent, Domain 3 mean runtime, and plan cost over time for Domain 2.

Theorems & Definitions (6)

  • Theorem 1
  • proof
  • Theorem 2
  • proof
  • Theorem 3
  • proof