BuildBB-0078Build of the week

Claude drives a 3-DOF tower crane through MCP, with a safety check in code

Amer El-Saad built a 3-DOF tower crane for his master's thesis that Claude plans and drives through 4 MCP tools. One overhead camera, YOLOv8 and ArUco markers locate parts and stations in centimeters. An Arduino Uno drives the DC motors, and code rejects any route that crosses a worker. README photos show the crane and live detections.

Open original source github.com · Amer El-Saad

Shown working
Why this evidence label?

README shows the physical crane setup, YOLOv8 and ArUco detections on the live overhead feed, before-and-after lens correction and the DC control-box schematic.

How we check entries
Preview image for Claude drives a 3-DOF tower crane through MCP, with a safety check in code
The source's own preview image, from github.com.

Briefing

Perception undistorts the camera image and maps pixels to table coordinates through a homography from 4 ArUco markers, accurate to about half a centimeter. The planner models obstacles as circles and tries two L-shaped routes; if both are blocked it halts and names the blocking object. Actuation sends timed motor pulses, and Claude issues intent through the tools on top. The model cannot override the check.

This repo is version 1. The author found it not accurate enough, because DC motors give no position feedback and position was estimated from run time. A separate Arduino Nano drives the electromagnet so a load stays held if the Uno resets. A rebuild with steppers, an ESP32 and a hook lives in a separate repo.

From the GitHub API

Repository facts

Repository facts for elsaadamer/ai-crane-automation, from the GitHub API
Repositoryelsaadamer/ai-crane-automation
Stars0
LanguagePython
Last push29 Sep 2026 30 d ago

Tags

Board
Agent
Interface

Report a problem with this entry