Skill Induction and Planning with Latent Language

TL;DR

This paper introduces a hierarchical policy learning framework that leverages sparse natural language annotations to discover reusable skills from demonstrations, enabling agents to plan with high-level instructions in complex environments.

Contribution

It presents a generative model that learns to parse demonstrations into high-level subtasks using minimal language annotations, improving interpretability and planning capabilities.

Findings

Achieves task completion rates comparable to state-of-the-art models.

Uses only 10% annotated demonstrations for training.

Provides structured, human-readable high-level plans.

Abstract

We present a framework for learning hierarchical policies from demonstrations, using sparse natural language annotations to guide the discovery of reusable skills for autonomous decision-making. We formulate a generative model of action sequences in which goals generate sequences of high-level subtask descriptions, and these descriptions generate sequences of low-level actions. We describe how to train this model using primarily unannotated demonstrations by parsing demonstrations into sequences of named high-level subtasks, using only a small number of seed annotations to ground language in action. In trained models, natural language commands index a combinatorial library of skills; agents can use these skills to plan by generating high-level instruction sequences tailored to novel goals. We evaluate this approach in the ALFRED household simulation environment, providing natural language annotations for only 10% of demonstrations. It achieves task completion rates comparable to state-of-the-art models (outperforming several recent methods with access to ground-truth plans during training and evaluation) while providing structured and human-readable high-level plans.