Schema helps AI agents learn how unknown environments work
Schema: Discovering Unknown Environments via Agentic Program Induction
Artificial Intelligence
Summary
Figuring out how to complete tasks in new places with unknown rules is hard for AI agents. This paper introduces Schema, a system that lets AI agents write small computer programs to explain and test how the world works as they learn. This helps the AI plan better and solve puzzles more accurately. The authors show that Schema greatly improves performance on several challenge tasks compared to previous methods.
What this means in practice
- •For game developers: Automatically generate AI characters that learn and adapt to new game environments with unknown rules.
- •For roboticists: Enable robots to discover and model the workings of unfamiliar environments to plan effective actions.
Authors
Guanning Zeng, Jiani Wang, Wenjie Ma, Shaofeng Yin, Chenyang Wang, Shichen Liu, Angjoo Kanazawa, Wode Ni, Xiuyu Li, Andrea Zanette, Haiwen Feng
Abstract
Learning to complete tasks in unfamiliar environments with unknown rules remains a key challenge for LLM agents. Current LLM agents often record their discoveries in prose, which may not provide a compact, explicit account of how the environment works. Inspired by how scientists organize observations into testable, predictive theories, we introduce Schema, an agent harness that organizes learning and action through interactive program induction. The LLM agent decides what to investigate and how to act, expressing its evolving understanding of the environment as executable programs. The harness consists of a persistent program workspace and a small set of interfaces for checking these programs against the interaction history, planning within them, and executing plans under step-by-step verification. Schema raises ARC-AGI-3 RHAE from 58.7% to 99.2% with the same base model, solves 100% of the public DiG-bench games, and reaches the median performance of the top-50 human players on MazeBench. Extensive analysis shows the effectiveness of Schema in unknown mechanism discovery, and ablations confirm the contribution of each component.