Large Language Models (LLM) based agents have shown promise in autonomously
completing tasks across various domains, e.g., robotics, games, and web
navigation. However, these agents typically require elaborate design and expert
prompts to solve tasks in specific domains, which limits their adaptability. We
introduce AutoManual, a framework enabling LLM agents to autonomously build
their understanding through interaction and adapt to new environments.
AutoManual categorizes environmental knowledge into diverse rules and optimizes
them in an online fashion by two agents: 1) The Planner codes actionable plans
based on current rules for interacting with the environment. 2) The Builder
updates the rules through a well-structured rule system that facilitates online
rule management and essential detail retention. To mitigate hallucinations in
managing rules, we introduce \textit{case-conditioned prompting} strategy for
the Builder. Finally, the Formulator agent compiles these rules into a
comprehensive manual. The self-generated manual can not only improve the
adaptability but also guide the planning of smaller LLMs while being
human-readable. Given only one simple demonstration, AutoManual significantly
improves task success rates, achieving 97.4\% with GPT-4-turbo and 86.2\% with
GPT-3.5-turbo on ALFWorld benchmark tasks. The source code will be available
soon.

通过自动生成规则和提高适应性，AutoManual 框架使基于大型语言模型（LLM）的代理能够自主构建自身的理解并适应新的环境。在 ALFWorld 基准任务上，通过 GPT-4-turbo 和 GPT-3.5-turbo，AutoManual 显著提高了任务成功率，并生成了人可读的综合手册。

AutoManual: 通过互动环境学习，由 LLM 代理生成指南手册

AutoManual: Generating Instruction Manuals by LLM Agents via Interactive  Environmental Learning

Using rules for home automation presents several challenges, especially when
considering multiple stakeholders in addition to residents, such as homeowners,
local authorities, energy suppliers, and system providers, who will wish to
contribute rules to safeguard their interests. Managing rules from various
sources requires a structured procedure, a relevant policy, and a designated
authority to ensure authorized and correct contributions and address potential
conflicts. In addition, the smart home rule language needs to express
conditions and decisions at a high level of abstraction without specifying
implementation details such as interfaces, access protocols, and room layout.
Decoupling high-level decisions from these details supports the transferability
and adaptability of rules to similar homes. This separation also has important
implications for structuring the smart home system and the security
architecture. Our proposed approach and system implementation introduce a rule
management process, a rule administrator, and a domain-specific rule language
to address these challenges. In addition, the system provides a learning
process that observes residents, detects behavior patterns, and derives rules
which are then presented as recommendations to the system.

利用规则来实现家庭自动化面临许多挑战，本文提出了一种规则管理过程、规则管理员和领域特定规则语言的系统实现来解决这些挑战，并提供了一个学习过程，观察居民的行为模式并推导规则作为对系统的推荐。