neodyd/robocopilot-x

RoboCopilot-X is an open-source AI assistant built for robotics developers, especially people working with ROS2 (a popular framework for programming robots) and the wider stack of tools around it. The pitch is that you type plain language requests, like "create a subscriber that logs the closest range from a laser scanner," and the tool turns that into runnable robotics code, project files, and configuration. The project frames itself as something like a code-completion sidekick that specifically understands robot-shaped problems instead of generic web or app code. The README lists eleven main features that are marked stable. These include a generator that produces complete ROS2 packages with the usual Python files and manifests, a composer for launch files that start several robot nodes together, scaffolders for MoveIt2 (motion planning), Nav2 (navigation behavior trees), URDF (robot body descriptions), and Isaac Sim (a robot simulator from Nvidia). It also includes a chat-style workspace UI with a code editor and file tree, an explainer that summarizes existing robotics codebases, and an agent system built on LangGraph that plans, writes, and checks code in a loop. Two more features are listed as in progress or planned: simulator-based testing of generated code, and adapters for newer vision language action robot policies. Architecturally, the project is a Next.js web frontend talking over HTTP and server-sent events to a FastAPI Python backend. The backend runs the agent orchestrator, which calls into separate generator modules that use templates to produce the actual ROS2, MoveIt2, Nav2, and Isaac Sim files. It is set up to work with any OpenAI-compatible language model service, so users can plug in OpenAI, Azure, a local Ollama install, vLLM, Together, DeepSeek, or Qwen. To run it, you clone the repo, copy the example environment file, add an API key for whichever model service you use, and start the stack with Docker Compose, which gives you the web UI on port 3000 and the API on port 8000. A local development path with separate Python and pnpm setup is also documented. The README includes a benchmark table claiming roughly 37x to 72x speed-ups against an internal intern cohort for tasks like wiring up a sensor subscriber or sketching a pick-and-place plan. The license is Apache 2.0.