LLM+P empowers large language models (LLMs) with optimal planning capabilities by translating natural language task descriptions into PDDL (Planning Domain Definition Language) for execution by classical planners. This project is targeted at researchers and developers working on AI planning, natural language understanding, and intelligent agents. It offers a novel approach to bridge the gap between human-readable instructions and formal planning representations.

How It Works

The core innovation lies in using LLMs to generate PDDL problem files from natural language descriptions. The system supports multiple methods: directly generating a plan from natural language, generating PDDL from natural language and then planning, or generating PDDL with contextual information for improved planning. This approach leverages the LLM's understanding of language to automate the creation of formal planning inputs, a traditionally manual and error-prone process.

Quick Start & Requirements

• Install: Requires OpenAI GPT API access and the Fast Downward planner.
• Prerequisites: OpenAI API keys must be placed in the keys/openai_keys.txt file.
• Running: Execute python main.py --domain DOMAIN --method METHOD --task TASK_ID or use bash run.sh DOMAIN METHOD TASK_ID.
• Supported Domains: barman, blocksworld, floortile, grippers, storage, termes, tyreworld.
• Supported Methods: llm_ic_pddl_planner, llm_pddl_planner, llm_planner, llm_ic_planner.
• Documentation: Fast Downward Official GitHub and website.

Highlighted Details

• Integrates LLMs with classical AI planning.
• Supports multiple methods for plan generation from natural language.
• Includes baseline comparisons for evaluation.
• Organizes code and data by domain and method.

Maintenance & Community

The project is associated with a pre-print: Liu et al., "LLM+P: Empowering Large Language Models with Optimal Planning Proficiency." Further community or maintenance information is not detailed in the README.

Licensing & Compatibility

The license is not specified in the README. Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

The project requires access to the OpenAI API, which may incur costs. The Fast Downward planner is a significant external dependency. The README does not specify the exact LLM models used or provide performance benchmarks.