Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sequential Discrete Action Selection via Blocking Conditions and Resolutions

Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita

TL;DR

This work presents a first instantiation of a strategy that frames the sequential action selection problem for robots in terms of resolving blocking conditions, i.e., situations that impede progress on an action en route to a goal.

Abstract

In this work, we introduce a strategy that frames the sequential action selection problem for robots in terms of resolving \textit{blocking conditions}, i.e., situations that impede progress on an action en route to a goal. This strategy allows a robot to make one-at-a-time decisions that take in pertinent contextual information and swiftly adapt and react to current situations. We present a first instantiation of this strategy that combines a state-transition graph and a zero-shot Large Language Model (LLM). The state-transition graph tracks which previously attempted actions are currently blocked and which candidate actions may resolve existing blocking conditions. This information from the state-transition graph is used to automatically generate a prompt for the LLM, which then uses the given context and set of possible actions to select a single action to try next. This selection process is iterative, with each chosen and executed action further refining the state-transition graph, continuing until the agent either fulfills the goal or encounters a termination condition. We demonstrate the effectiveness of our approach by comparing it to various LLM and traditional task-planning methods in a testbed of simulation experiments. We discuss the implications of our work based on our results.

Sequential Discrete Action Selection via Blocking Conditions and Resolutions

TL;DR

This work presents a first instantiation of a strategy that frames the sequential action selection problem for robots in terms of resolving blocking conditions, i.e., situations that impede progress on an action en route to a goal.

Abstract

In this work, we introduce a strategy that frames the sequential action selection problem for robots in terms of resolving \textit{blocking conditions}, i.e., situations that impede progress on an action en route to a goal. This strategy allows a robot to make one-at-a-time decisions that take in pertinent contextual information and swiftly adapt and react to current situations. We present a first instantiation of this strategy that combines a state-transition graph and a zero-shot Large Language Model (LLM). The state-transition graph tracks which previously attempted actions are currently blocked and which candidate actions may resolve existing blocking conditions. This information from the state-transition graph is used to automatically generate a prompt for the LLM, which then uses the given context and set of possible actions to select a single action to try next. This selection process is iterative, with each chosen and executed action further refining the state-transition graph, continuing until the agent either fulfills the goal or encounters a termination condition. We demonstrate the effectiveness of our approach by comparing it to various LLM and traditional task-planning methods in a testbed of simulation experiments. We discuss the implications of our work based on our results.
Paper Structure (24 sections, 2 figures, 3 tables)

This paper contains 24 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Task Planning
  4. LLMs for Planning
  5. Strategy Overview
  6. Strategy Components
  7. Strategy Steps
  8. Strategy Example
  9. Initial Instantiation of Strategy
  10. Selection Engine using LLM
  11. LLM Prompt Generation
  12. Evaluation
  13. Implementation Details
  14. Experimental Testbed
  15. Baseline Comparisons
  16. ...and 9 more sections

Figures (2)

  • Figure 1: We present a strategy that frames action selection in terms of resolving blocking conditions, i.e., situations that impede progress on an action en route to a goal. This example illustrates a robot using blocking conditions and resolutions to complete the task "wash the cup in the sink" in the AI2Thor simulation environment. This task is part of our evaluation.
  • Figure 2: The tree structure during the retrieving milk from the refrigerator example. The red nodes signify blocked actions, the orange diamonds signify blocking conditions, gray nodes signify available actions, the green node is the currently selected action, and the blue symbols signify resolution actions that may fix the parent blocking condition.