Connecting to sources…

How to build RoBart LLM-controlled iPhone and hoverboard robot: the whole process

20 steps to build RoBart LLM-controlled iPhone and hoverboard robot, from buying parts, printing and assembly to software, calibration and training, with the bill of mat…

The creator gives no build cost or BOM total; the README only says the aim was a cheap mobile base from a hoverboard and an iPhone Pro. An iPhone Pro is needed too, plus LLM and Deepgram API usage

Print and make

  • Print the phone mount and charger case

    Use the STL files in the parts/ directory to print the phone mount and a case that houses one original Hover-1 H1 logic board for use as a battery charger.

Assemble

  • Build the RoBart frame

    Use 2020 aluminum extrusions to create the frame, and attach the rear caster and iPhone to it.

  • Install the phone holder on the upright

    Mount the 3D-printed phone holder atop the vertical aluminum extrusion and attach the phone holder.

  • Add the removable frame connector

    Cut and join the extrusion with a removable plastic connector so RoBart can be transported in a car trunk.

  • Connect the motors, controllers, and battery

    Use suitable connectors to connect the motors to the RioRand driver boards, connect the driver boards to the Feather with Dupont connectors, and route battery power with XT60 splitter cables.

  • Assemble the battery charger case

    Place one original Hover-1 H1 logic board inside the 3D-printed case and use the assembly as a battery charger.

Software

  • Use AnnotatingCamera to Annotate Photos

    Have the camera take photos and annotate them with navigable points using the occupancy map.

  • Run the Brain Speech Task Agent

    Start the Brain so it listens for speech and attempts the requested task using all of RoBart's capabilities and the chosen LLM.

  • Connect the Motor Controller via BLE

    Use HoverboardController on iOS to communicate with the Arduino firmware via BLE and send individual motor throttle values ranging from -1.0 to +1.0.

  • Enable iOS microphone transcription

    Provide a Deepgram API key in the settings view. The iOS app uses WebRTC VAD to detect human speech and uploads audio to Deepgram for transcription.

Calibrate and test

  • Verify the Motor Power Watchdog

    Verify that the watchdog cuts motor power if the BLE connection is lost or throttle values are not updated within a certain number of seconds. PID-controlled modes send constant updates to prevent the watchdog from engaging.

Data and training

  • Annotate ARKit photos with navigable points

    Have AnnotatingCamera capture ARKit photos containing camera pose and intrinsics, and use the floor elevation provided by ARSessionManager to place fixed points near the robot.

  • Stream audio from Apple Watch

    Use the Watch app target to make Apple Watch act as a microphone and stream audio seamlessly to the iOS app through Watch Connectivity. Start and stop recording manually for more reliable capture.

Use it

  • Set Open-Loop Left and Right Throttles

    Use drive to set the left and right motor throttle values independently in open loop.

  • Rotate in Place by Angular Velocity

    Use rotateInPlace to turn in place at the given angular velocity; -1 means left at full throttle and +1 means right at full throttle, using open-loop control.

  • Rotate in Place by Specified Degrees

    Use rotateInPlaceBy to rotate the robot in place by the specified number of degrees.

  • Face a World-Space Direction

    Use face to turn in place until the robot faces the given world-space direction vector.

  • Drive Forward a Specified Distance

    Use driveForward to drive the robot forward by the specified distance in meters.

  • Drive to a World-Space Position

    Use driveTo to turn toward the target and drive in a straight line to the specified world-space position.

  • Drive While Maintaining a Direction

    Use driveToFacing to drive to the specified position while facing the given direction; it is particularly useful for driving backwards to a point.

Parts

  • A Hover-1 H1 hoverboard with its shell and original electronics removed, keeping the hub motors and 36V battery; one original logic board goes into a printed case as the charger
  • A 2020 aluminium-extrusion frame, a rear caster and a vertical mast joined by a removable printed connector
  • An Adafruit Feather nRF52 Bluefruit LE, a DC-DC converter from 36V to 3.3V, two RioRand 350W brushless drivers, Dupont connectors and XT60 splitters
  • Printed parts in parts/: 0° and 15° phone mounts, the case top and bottom, extrusion connector and end caps, and hoverboard circuit-board mounting plates
  • An iPhone Pro (the README relies on its LiDAR), optionally an Apple Watch

Tools

  • A 3D printer, and a way to cut aluminium extrusion (the mast is cut and rejoined with a printed connector)
  • A Mac with Xcode (to build and install the iOS app from source), and the Arduino IDE (to flash the Feather)

Skills

  • Stripping a hoverboard and wiring the brushless drivers and the Feather
  • Swift and iOS development (ARKit, RealityKit, Metal), and configuring the model and API keys in settings
  • Optional: issuing debug commands through the Python debug server

Software

  • The RoBart iOS app (GPL‑3.0): ARSessionManager (ARKit), speech detection (WebRTC VAD plus Deepgram), HoverboardController (PID trajectories), NavigationController (occupancy map, crude pathfinding, following the nearest person), AnnotatingCamera and the ReAct-like agent Brain
  • The Feather's Arduino firmware (hoverboard/), the watchOS remote app and the Python debug server (server/); the web teleoperation client in the webrtc-teleop branch

Documents