The idea: teach a robot by demonstration instead of programming it - show, correct, judge → imitation learning + reward model + RL, with the operator never touching the ML.

Live today (early, honest): open mobile base with autonomous nav (ROS 2 + Nav2 + RPLIDAR SLAM), no-code Blockly UI with live telemetry + mapping, and an LLM voice-command layer. Two-layer architecture: Raspberry Pi 5 (high-level: ROS 2, SLAM, nav, UI) + Teensy 4.0 (real-time: motors, encoders, IMU). Differential drive, 24V BLDC motors. Documented, MIT-licensed.

Next: harden electronics (custom PCB replacing the breadboard, safety layer, power distribution), then start the dual-arm manipulation + learn-by-demonstration pipeline (foundation-model fine-tuning + simulation + human-in-the-loop RL). The mobile base is the on-ramp; the learning platform is the real project.

Rough in places, built in public. Feedback and contributors very welcome.
Repo: github.com/openAMRobot