OM1

Why we included this project

Robotics teams that want a robot to reason with language models often end up wiring up every hardware driver and LLM endpoint by hand. OM1 is a hardware abstraction layer and agent runtime that does that integration work for you: one multimodal pipeline drives different robot bodies, from a TurtleBot 4 to a simulated humanoid in Gazebo or Isaac Sim, and you can swap in whichever LLM or vision-language model you already use. The Go implementation targets edge constraints like lower latency and a smaller memory footprint, and it deploys as a single binary with Zenoh bundled. It also ships preconfigured endpoints for several providers and a Prometheus plus Grafana stack for tracking pipeline latencies, so you can see where time goes without building your own observability. Even if you prefer the deprecated Python branch, the modular plugin design is worth studying as a clean pattern for keeping robot software portable across hardware.