Opus 5.5 Designed This Robot. I Just Built It. (And It WORKS!)

Highlights

• A 5-servo robotic arm named Greg performs high-fives using an ESP32 S3 board, a secondary ESP32 camera board, and Google MediaPipe running on a Mac server.

• Claude Opus 5.5 generated all the 3D models in Blender, wiring diagrams, firmware for both boards, and computer vision algorithms across a single month of iteration.

• A 1000 microfarad capacitor connected across the 5-volt and ground lines prevents voltage dips from resetting the ESP32 when multiple heavy servos start moving simultaneously.

• A 120-degree wide-angle camera module combined with a 25-degree wedge on a custom camera box resolves narrow field-of-view issues during hand detection.

• Voice control integration and 268 calibration poses enable the robot to learn spatial mapping and match wrist angles to a human hand independently.

Timeline

1. 00:00 - 01:48

   Hardware architecture and initial concept

   The project serves a single purpose: responding to an raised hand with a high-five gesture. Video feeds stream over Wi-Fi from a secondary camera board to a Mac server running computer vision detection. Hardware coordination runs through a single power supply to streamline startup procedures.

   • A robotic high-five system operates via five servos controlled by an ESP32 S3 board and an ESP32 camera module.
   • Google MediaPipe running on a Mac server detects open palm gestures and calculates joint angles for the arm.
   • Design and mechanical components are offloaded entirely to Claude Opus 5.5 to bypass traditional CAD tools.
2. 01:48 - 04:44

   3D design iteration and prototyping with Blender

   Materials include an Elegoo Neptune 3 Pro 3D printer alongside spare parts and an ESP32 camera. Physical iteration involves photographing misfits, sending them to the AI, and generating updated 3D files. Power distribution relies on standard DuPont cables and pin connectors linked to a custom power hub.

   • Blender serves as the 3D design software driven by Python scripts generated by Claude Opus.
   • Small fit tests for servo pockets prevent filament waste before executing full 8-to-10-hour base prints.
   • A custom perf board inside the electronics bay functions as a power hub using soldered 5-volt and ground lines.
3. 04:44 - 07:16

   Circuit stability, wiring rules, and software bug fixes

   A 5-volt 6-amp supply powers the system, where sudden movements by MG995 servos cause temporary voltage dips. Power must route strictly through the board rather than the ESP32 pin. Initial flaws, such as missing wire dimensions in AI-generated shoulder animations, highlight the necessity of visual verification before final assembly.

   • A 1000 microfarad capacitor prevents voltage drops from resetting the microcontroller during heavy servo movement.
   • Shared ground lines prevent servo trembling by providing a consistent control signal reference.
   • Manual PWM channel assignment in firmware resolves overlapping joint signals generated by default library configurations.
4. 07:16 - 10:20

   Vision upgrades, camera box design, and calibration

   Early camera mounts suffer from incorrect thickness and antenna placement assumptions. Version 3 features a secure camera box lid and a 25-degree wedge pointing upwards. Calibration involves moving through 268 poses while tracking tape on the robot palm to generate a mapping model for joint angles.

   • A 120-degree wide-angle camera module replaces the original narrow lens to capture various hand-raise locations.
   • A fifth joint twist provides door-knob-like rotation so the palm matches the angle of the user's hand.
   • A voice control page allows real-time adjustment of joint angles and wrist rotation during calibration.
5. 10:20 - 12:51

   Autonomous deployment and reaction speed optimization

   Final assembly mounts the base onto a cutting board for stability against swinging momentum. Sliders on a custom web page enable tuning for delay, speed, and slap depth without reflashing firmware. All source code, parts lists, and design files are published in a GitHub repository.

   • Independent Wi-Fi firmware configurations on both boards allow automatic connection to the Mac upon startup.
   • Status lights transitioning from red to yellow, blue, and green indicate active boot and network connection phases.
   • Reaction delays drop from one second to a quarter of a second with tripled servo speed limits.