Skip to content
Solan Homestead
← projects

computer vision + hardware

AI Grant Trashcan

A colleague and I received a $5,000 Clark AI Innovation Fund grant to build a trashcan that uses computer vision to identify waste and route it with a physical sorting platform.

in development · $5,000 grant

The Raspberry Pi camera feeds a three-class YOLO model, confirms an object over multiple frames, and sends an acknowledged command to an ESP32 that controls the sorter.

The current mechanism uses one 270-degree positional servo. Trash and recycling move the same platform in opposite directions and then return it to the midpoint, with a local control panel for calibration and supervised movement tests.

Numbers that matter

  • $5,000 Clark AI Innovation Fund grant
  • 3 waste classes
  • single 270° sorting platform

How it fits together

  1. 01Pi camera
  2. 02YOLO26n inference
  3. 03multi-frame object confirmation
  4. 04serial command queue
  5. 05ESP32 + 270° platform servo

What I learned

  • The physical mechanism creates failure modes that a model benchmark does not show.
  • Manual calibration and actuator state need to share the same control path as automatic sorting.