Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

POrTAL: Plan-Orchestrated Tree Assembly for Lookahead

Evan Conway, David Porfirio, David Chan, Mark Roberts, Laura M. Hiatt

TL;DR

<3-5 sentence high-level summary>Planners for robots must operate under uncertainty with limited compute, which makes efficient online planning crucial. The authors introduce POrTAL, a lightweight anytime planner that blends FF-Replan's determinization with POMCP-style lookahead to generate shorter plans while remaining computationally tractable. In two domains (office and elevator) with uncertain item locations, POrTAL often outperforms FF-Replan and POMCP, especially under skewed likelihoods, and demonstrates favorable performance with modest planning time budgets. The work highlights the practical potential of hybrid planning strategies for real-time robotic decision-making under uncertainty, while also identifying limitations and directions for real-world deployment and broader scenario coverage.

Abstract

Assigning tasks to robots often involves supplying the robot with an overarching goal, such as through natural language, and then relying on the robot to uncover and execute a plan to achieve that goal. In many settings common to human-robot interaction, however, the world is only partially observable to the robot, requiring that it create plans under uncertainty. Although many probabilistic planning algorithms exist for this purpose, these algorithms can be inefficient if executed with the robot's limited computational resources, or may require more steps than expected to achieve the goal. We thereby created a new, lightweight, probabilistic planning algorithm, Plan-Orchestrated Tree Assembly for Lookahead (POrTAL), that combines the strengths of two baseline planning algorithms, FF-Replan and POMCP. In a series of case studies, we demonstrate POrTAL's ability to quickly arrive at solutions that outperform these baselines in terms of number of steps. We additionally demonstrate how POrTAL performs under varying temporal constraints.

POrTAL: Plan-Orchestrated Tree Assembly for Lookahead

TL;DR

<3-5 sentence high-level summary>Planners for robots must operate under uncertainty with limited compute, which makes efficient online planning crucial. The authors introduce POrTAL, a lightweight anytime planner that blends FF-Replan's determinization with POMCP-style lookahead to generate shorter plans while remaining computationally tractable. In two domains (office and elevator) with uncertain item locations, POrTAL often outperforms FF-Replan and POMCP, especially under skewed likelihoods, and demonstrates favorable performance with modest planning time budgets. The work highlights the practical potential of hybrid planning strategies for real-time robotic decision-making under uncertainty, while also identifying limitations and directions for real-world deployment and broader scenario coverage.

Abstract

Assigning tasks to robots often involves supplying the robot with an overarching goal, such as through natural language, and then relying on the robot to uncover and execute a plan to achieve that goal. In many settings common to human-robot interaction, however, the world is only partially observable to the robot, requiring that it create plans under uncertainty. Although many probabilistic planning algorithms exist for this purpose, these algorithms can be inefficient if executed with the robot's limited computational resources, or may require more steps than expected to achieve the goal. We thereby created a new, lightweight, probabilistic planning algorithm, Plan-Orchestrated Tree Assembly for Lookahead (POrTAL), that combines the strengths of two baseline planning algorithms, FF-Replan and POMCP. In a series of case studies, we demonstrate POrTAL's ability to quickly arrive at solutions that outperform these baselines in terms of number of steps. We additionally demonstrate how POrTAL performs under varying temporal constraints.

Paper Structure

This paper contains 29 sections, 3 figures, 3 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Automated Planning
  4. Probabilistic Planning
  5. FF-Replan
  6. POMCP
  7. Other approaches
  8. Technical Approach
  9. Preliminaries
  10. FF-Replan
  11. POMCP
  12. POrTAL
  13. Evaluation
  14. Domains
  15. Office
  16. ...and 14 more sections

Figures (3)

  • Figure 1: Simple household environment in which a robot is tasked to deliver a cup to the kitchen table. The cup has a 20% chance of being in the kitchen already, and an 80% chance of being in the bathroom. Although there is a smaller chance of the cup being in the kitchen, a policy that minimizes the number of anticipated steps to deliver the cup may dictate that the robot check in the kitchen first.
  • Figure 2: A visual depiction of both domains used in our evaluation. The office domain (top) consists of an office environment with rooms (white) and hallways (grey). Red nodes are region cells and blue nodes are surface cells. Adjacent cells are linked based on the robot's ability to travel between them. The elevator domain consists of two floors of a building linked by an elevator. Each floor is divided into a 5x4 grid of region cells that the robot can travel between. Some region cells have surface cells inside of them.
  • Figure 3: The results of our Algorithm Comparison (left) and Time Comparison (right). Error bands represent standard error of the mean.