Skip to content

Robot's identity within RAI

RAI Agent needs to understand what kind of robot it is running on. This includes its looks, purpose, ethical code, equipment, capabilities and documentation. To configure RAI for your robot, provide contents for your robot's so called whoami package.

Configuration example - Franka Emika Panda arm

  1. Setup the repository using quick setup guide

  2. Fill in the panda/ folder with data:

    2.1. Save this image into panda/images

    2.2. Save this document in panda/documentation

    2.3. Save this urdf in panda/urdf

  3. Build the embodiment info using build_whoami.py:

    python src/rai_whoami/rai_whoami/build_whoami.py panda/ --build-vector-db

    Vector database

    Building the vector database with cloud vendors might lead to costs. Consider using the local ollama provider for this task. The embedding model can be configured in config.toml (ollama works locally, see docs/setup/vendors.md).

  4. Examine the generated files

After running the build command, inspect the generated files in the panda/generated directory. The folder should contain a info.json file containing:

  • rules: List of rules
  • capabilities: List of capabilities
  • behaviors: List of behaviors
  • description: Description of the robot
  • images: Base64 encoded images

Testing (ROS 2)

You can test the generated package by using the RAI Whoami services:

  1. Using the RAI Whoami services:

Run the RAI Whoami services:

python src/rai_whoami/rai_whoami/scripts/ros2_embodiment_service.py panda/ &
python src/rai_whoami/rai_whoami/scripts/ros2_vector_store_retrieval_service.py panda/

With the services running, you can query the robot's identity and vector database:

# Get robot's identity
ros2 service call /rai_whoami_embodiment_info_service rai_interfaces/srv/EmbodimentInfo

# Query the vector database
ros2 service call /rai_whoami_documentation_service rai_interfaces/srv/VectorStoreRetrieval "query: 'maximum load'"

If your service calls succeed and you can access the embodiment info and vector database, your robot's whoami package has been properly initialized.

  1. Alternatively, you can use the RAI Whoami tools directly in your Python code:
from rai_whoami import EmbodimentInfo
from rai_whoami.tools import QueryDatabaseTool

# Load embodiment info
info = EmbodimentInfo.from_directory("panda/generated")

# Create a system prompt for your LLM
system_prompt = info.to_langchain()

# Use the vector database tool
query_tool = QueryDatabaseTool(root_dir="panda/generated")
query_tool._run(query="maximum load")