Vague, but exciting…

Robotics

Meet the Robo Bump Truck

I’m building a small RC truck that can follow a trail using camera images. This original film introduces the idea: use richer sensors to teach a model, then see what it can learn to recognize from a camera alone.

Watch the original film on YouTube · Read Log #1: the kitchen experiment.

Where it began — August 2026. The film covers the early obstacle-learning experiment. Since then, the work has moved into trail following, repeatable kitchen laps, and understanding where camera-based steering succeeds or fails. The latest update is below; the full story follows in order.

Latest: getting the truck following the line again

Read Log #4 or watch the film. An October 2 setback, three changes made together, and better line-following later that night. The comparison shows progress without establishing which change helped. The next chapter heads out onto real trails.

Earlier: kitchen laps

Read Log #3: My robot lapped my kitchen on its own or watch the film. Camera timing, steering geometry, and measurement limits are part of the story. The article describes the conditions tested then; it is not a statement of the truck's current readiness.

Follow the story

  1. Log #1 — learning what to avoid. Read the kitchen experiment · Watch. Other sensors supply training labels; the resulting model takes camera images.
  2. Log #2 — a trail and a difficult corner. Read the field log · Watch the released short, “It drove itself”. The article explains the division between model steering and human throttle control.
  3. Log #3 — a kitchen racetrack. Read the indoor log · Watch. A repeatable course helps expose faults in the sensing and steering pipeline.
  4. Log #4 — getting the line-following back. Read the comparison · Watch. A failed corner and a later useful stretch, with three simultaneous changes and an unresolved cause.

Robot Eyes: the iPhone sensor app

Robot Eyes: Vision & Pose is available on the App Store. I built it to record camera video, LiDAR depth, motion sensors and ARKit pose for robotics experiments and dataset collection. It also supports a live camera stream to a connected computer.

The published app and our development builds are separate. The App Store listing currently shows version 1.1; newer experimental features are not implied by that release. The app source repository remains private.

Build and troubleshooting references