Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Hierarchical Domain Models Through Environment-Grounded Interaction

Claudius Kienle, Benjamin Alt, Oleg Arenz, Jan Peters

TL;DR

LODGE proposes autonomous, environment-grounded learning of hierarchical domain models to enable open-world planning for long-horizon tasks. By combining LLM-generated abstractions with environment grounding, hierarchical decomposition, predicate invention, and learned predicate classifiers, LODGE continually refines domain models through motion verification and a global recovery module, all without human annotations. Across IPC domains and a robotic FurnitureBench task, LODGE yields more accurate domain models and higher task success with far fewer environment interactions than prior methods. The approach demonstrates strong potential for autonomous open-world reasoning by distributing complexity across hierarchy levels and grounding symbolic state in continuous perception.

Abstract

Domain models enable autonomous agents to solve long-horizon tasks by producing interpretable plans. However, in open-world environments, a single general domain model cannot capture the variety of tasks, so agents must generate suitable task-specific models on the fly. Large Language Models (LLMs), with their implicit common knowledge, can generate such domains, but suffer from high error rates that limit their applicability. Hence, related work relies on extensive human feed-back or prior knowledge, which undermines autonomous, open-world deployment. In this work, we propose LODGE, a framework for autonomous domain learning from LLMs and environment grounding. LODGE builds on hierarchical abstractions and automated simulations to identify and correct inconsistencies between abstraction layers and between the model and environment. Our framework is task-agnostic, as it generates predicates, operators, and their preconditions and effects, while only assuming access to a simulator and a set of generic, executable low-level skills. Experiments on two International Planning Competition ( IPC) domains and a robotic assembly domain show that LODGE yields more accurate domain models and higher task success than existing methods, requiring remarkably few environment interactions and no human feedback or demonstrations.

Learning Hierarchical Domain Models Through Environment-Grounded Interaction

TL;DR

LODGE proposes autonomous, environment-grounded learning of hierarchical domain models to enable open-world planning for long-horizon tasks. By combining LLM-generated abstractions with environment grounding, hierarchical decomposition, predicate invention, and learned predicate classifiers, LODGE continually refines domain models through motion verification and a global recovery module, all without human annotations. Across IPC domains and a robotic FurnitureBench task, LODGE yields more accurate domain models and higher task success with far fewer environment interactions than prior methods. The approach demonstrates strong potential for autonomous open-world reasoning by distributing complexity across hierarchy levels and grounding symbolic state in continuous perception.

Abstract

Domain models enable autonomous agents to solve long-horizon tasks by producing interpretable plans. However, in open-world environments, a single general domain model cannot capture the variety of tasks, so agents must generate suitable task-specific models on the fly. Large Language Models (LLMs), with their implicit common knowledge, can generate such domains, but suffer from high error rates that limit their applicability. Hence, related work relies on extensive human feed-back or prior knowledge, which undermines autonomous, open-world deployment. In this work, we propose LODGE, a framework for autonomous domain learning from LLMs and environment grounding. LODGE builds on hierarchical abstractions and automated simulations to identify and correct inconsistencies between abstraction layers and between the model and environment. Our framework is task-agnostic, as it generates predicates, operators, and their preconditions and effects, while only assuming access to a simulator and a set of generic, executable low-level skills. Experiments on two International Planning Competition ( IPC) domains and a robotic assembly domain show that LODGE yields more accurate domain models and higher task success than existing methods, requiring remarkably few environment interactions and no human feedback or demonstrations.
Paper Structure (49 sections, 4 equations, 2 figures, 5 tables, 1 algorithm)

This paper contains 49 sections, 4 equations, 2 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Planning with llm
  4. Learning Domain Models with llm
  5. Predicate Invention and Learning Classifiers
  6. Notation and Problem Statement
  7. METHODS
  8. Learning Hierarchical Domain Models by Planning
  9. Abstraction through hierarchical operators
  10. Plan generation under partial domain models
  11. Mapping to skills
  12. Decomposition
  13. Aligning Levels of Hierarchical Domain Models
  14. Inventing Predicates and Learning Classifiers
  15. Data Collection
  16. ...and 34 more sections

Figures (2)

  • Figure 1: LODGE: Learning hierarchical domain models with environment grounding.
  • Figure 2: The hierarchical planning of assembling the lamp with LODGE for the FurnitureBench environment, including decomposition of pick-up(bulb) and re-planning within screw-in(bulb).