gpt_umrf_parser

GPT-based natural language to UMRF parser. This package contains the code for replicating the code in the journal paper:

Unlocking Underrepresented Use-Cases for Large Language Model-Driven Human-Robot Task Planning.

The repository is broken into three main branches:

• main - provides quick and simple scripts for querying OpenAI API with natural language for UMRF task and motion planning.
• selma-devel - provides the datasets, LLM code for huggingface API and OpenAI API, and prompt builder to perform the prompt experiments in Section 3.
• demo_test - provides the ROS code to perform the remote inspection demonstration in Section 5.

Please refer to each branch's README.md file for full setup instructions.

Main Branch Setup

• OpenAI API requires Numpy v1.20.3 or higher, so if that's not the case then: sudo pip install numpy --upgrade
• Install openai python package: pip install openai.
• Generate a key for using openai API.
• Store this key in a separate file.
• Clone this project git clone https://github.com/temoto-framework/gpt_umrf_parser

Usage

Standalone

# Export the key as an environment variable
export GPT_API_KEY=$(cat <path/to/openai_key>)

# Invoke the script
python scripts/gpt_umrf_parser_standalone.py -ue umrf_examples/ -is "Robot go scan the lab [x=111.2; y=87.6; yaw=-0.11]."

ROS node

# Export the key as an environment variable
export GPT_API_KEY=$(cat <path/to/openai_key>)

# Invoke the node
rosrun gpt_umrf_parser gpt_umrf_parser_node.py -ue umrf_examples/

# Publish the command
rostopic pub /command std_msgs/String "data: 'Robot go scan the lab [x=111.2; y=87.6; yaw=-0.11].'"

# Subscribe to the result
rostopic echo /broadcast_start_umrf_graph