Automatically generated curricula help learn task hierarchies efficiently
Curriculum-Guided Acquisition of Hierarchical Task Knowledge from LLM-Generated Experience
Artificial Intelligence
Summary
Solving problems often involves reusing skills learned from similar tasks, but teaching a computer these skills can be complex and expensive. The authors study ways to guide computers to learn these skills more effectively by creating a sequence of simpler learning tasks, or curricula. They compare learning while solving problems, using expert-crafted curricula, and automatically making curricula from solutions generated by large language models. Their experiments show that the automatic process nearly matches expert designs while greatly reducing the cost of acquiring these skills.
What this means in practice
- •For robotics developers: Train robots to learn complex task sequences more efficiently using automated curricula derived from AI-generated examples.
- •For game ai programmers: Improve the development of non-player characters by accelerating the acquisition of hierarchical task methods from AI-simulated gameplay data.
Authors
Yuesheng Xu, Hector Munoz-Avila
Abstract
Hierarchical task knowledge enables efficient problem solving by allowing previously acquired skills to be reused across tasks and problem instances. In this paper, we investigate curriculum-guided acquisition of hierarchical task knowledge in an HTN learning framework. We compare three approaches: online learning during problem solving, human-designed curricula, and an automated curriculum generation procedure that derives learning experiences from verified LLM-generated solution traces. The proposed procedure solves small training problems with an LLM, extracts task accomplishment information, and transforms it into ordered sequences of state-task learning problems for HTN method acquisition. We evaluate the three approaches across multiple planning domains. The results demonstrate that automatically generated curricula recover much of the benefit of expert-designed curricula while substantially reducing the token cost of online hierarchical knowledge acquisition.