Connecting to sources…

RoBart LLM-controlled iPhone and hoverboard robot · A mobile base made from a salvaged hoverboard: a Hover-1 H1 with its plastic shell and original electronics removed, a 2020 aluminium-extrusion frame with a rear caster, and a 3D-printed holder atop a vertical extrusion carrying an iPhone Pro for compute and sensors; the hoverboard's 36V battery powers two RioRand 350W brushless drivers for the hub motors, and an Adafruit Feather nRF52 takes throttle commands from the iPhone over Bluetooth LE; the mast splits at a removable connector so the robot fits in a car trunk. The creator gives no size or weight

A mobile base made from a salvaged hoverboard: a Hover-1 H1 with its plastic shell and original electronics removed, a 2020 aluminium-extrusion frame with a re…

Maker
Bart Trzynadlowski (trzy) open-source project
Form
A mobile base made from a salvaged hoverboard: a Hover-1 H1 with its plastic shell and original electronics removed, a 2020 aluminium-extrusion frame with a rear caster, and a 3D-printed holder atop a vertical extrusion carrying an iPhone Pro for compute and sensors; the hoverboard's 36V battery powers two RioRand 350W brushless drivers for the hub motors, and an Adafruit Feather nRF52 takes throttle commands from the iPhone over Bluetooth LE; the mast splits at a removable connector so the robot fits in a car trunk. The creator gives no size or weight
For
The creator says he open-sourced it to stimulate discussion and exploration (navigation, agents, an easier-to-replicate mechanical design, adding an arm, applications), and calls the mechanical design a one-off; the repo provides the iOS app source, firmware and printed parts, for developers and researchers who can build iOS apps and rework hardware
Sensors
All sensing comes from the iPhone Pro: RGB cameras, LiDAR, microphones and speakers; ARKit provides 6-DoF tracking (SLAM), scene meshing and floor detection, from which the app builds an occupancy map and marks reachable points on photos. The motors have no encoders and the Arduino side has no feedback; an optional Apple Watch app works as a push-to-talk microphone
Connectivity and app
The robot is the RoBart iOS app on the iPhone (with an optional watchOS app): it sends left and right motor throttles over Bluetooth LE to the Feather firmware and closes PID loops on ARKit's pose; the app needs no companion server and calls public LLM APIs directly (settings take Anthropic, OpenAI and Mistral keys), with Deepgram for speech transcription. An optional Python debug server offers an interactive terminal, and since 2026‑01 the webrtc-teleop branch adds WebRTC teleoperation from a web client
Subscription
Self-built open source with no subscription; the app's settings need your own LLM (Anthropic, OpenAI or Mistral) and Deepgram API keys
Openness
The whole repo is GPL‑3.0 (the LICENSE file opens with a one-line note, so GitHub reports NOASSERTION): the iOS and watchOS apps, the Feather's Arduino firmware, the Python debug server and the STL and 3MF files for every printed part under parts/. The README says the code for the contactless health-monitoring prototype is not included
Privacy
The iPhone's cameras, LiDAR and microphone are in constant use: speech is uploaded to Deepgram for transcription, and transcripts and photos go straight from the phone to the chosen LLM service (Claude, GPT or Mistral), with the API keys saved in the app settings; the teleoperation branch connects to a web client over WebRTC. The repo does not say how these cloud services retain data.
Not stated
Size, weight, speed, battery capacity and run time, payload, build cost or BOM total, the exact iPhone model needed (the README says only iPhone Pro), recommended age

Vendor claims

  • A cheap mobile base from a salvaged hoverboard and an iPhone Pro: the phone brings plenty of compute, connectivity, RGB cameras, LiDAR, microphones and speakers, and ARKit brings SLAM and scene understanding (README)
  • An LLM agent such as Claude or GPT‑4 runs it: it listens to speech, looks at photos and decides how to satisfy a request with a handful of basic actions such as turning, driving and taking photos, with no task-specific code (video description)
  • Applications the creator suggests exploring: accessibility, tidying up, contactless health monitoring, home inspection and security, a robotic porter (README)

Sources