I am pleased to announce that we made a new release on ROScribe that supports a major feature which comes very helpful in robot integration.
Training ROScribe on ROS index
We trained ROScribe on all open source repositories and ROS packages listed on ROS index. Under the hood, we load all documents and metadata associated with all repositories listed on ROS index into a vector database and use RAG (retrieval augmented generation) technique to access them. Using this method, we essentially teach the LLM (gpt3.5 in our default setting) everything on ROS Index to make it an AI agent expert in robotics.
ROScribe is trained on all ROS versions (ROS & ROS 2) and all distributions.
Use ROScribe as a robotics expert
With this release you can use ROScribe as your personal robotics consultant. You can ask him any technical question within robotics domain and have him show you the options you have within ROS index to build your robot. You can ask him to show you examples and demos of a particular solution, or help you install and run any of the ROS packages available in ROS index.
Here is a demo that shows ROScribe helping a robotics engineer to find a multilayer grid mapping solution and shows him how to install it.
To run ROScribe for this specific feature use: roscribe-rag in your command line.
You can find more info on our github and its wiki page.
New in this release
Here are what’s new in release v0.0.4:
- Scripts for automatic extraction of ROS package documentation given your choice of ROS version
- Build a vector database over ROS Index
Retrieval augmented generation (RAG) capabilities for ROScribe:
- Now ROScribe has access to the most recent open-source ROS repositories that can be found on ROS Index
- ROScribe can be called as an AI agent that assists you with finding the relevant ROS packages for your project
- Use roscribe-rag to run the RAG agent
Creating a wiki page for documentation to keep the readme file short.
About ROScribe & future roadmap
ROScribe is our AI-native robot integration solution.You can describe your robotic project in natural language and ROScribe generates the entire robot software for you (within ROS framework).
As of now, the entire code is generated by the LLM, meaning that the RAG feature (explained above) is currently a stand-alone feature and isn’t fully integrated into the main solution. We are working on a fully-integrated solution that retrieves the human-written (open source) ROS packages whenever possible (from ROS index or elsewhere), and only generates code when there is no better code available. This feature will be part of our next release.
We also plan to give ROScribe a web-based GUI.
Please checkout our github and let us know what you think.