{"slug": "claude-helped-build-a-robot-that-high-fives", "title": "Claude Helped Build a Robot That High-Fives", "summary": "A builder used Anthropic's Claude Opus 3.5 to generate the complete design package for \"Greg,\" a high-five robot built on an ESP32-S3 board with five servos, a secondary ESP32-CAM streaming video over Wi-Fi to a Mac running Google MediaPipe for palm detection. Claude produced the 3D models as Blender Python scripts, wiring diagrams, firmware for both ESP32 boards and the MediaPipe vision code, while the human printed parts on an Elegoo Neptune 3 Pro and iterated by feeding photos of fit problems back to the model. The build surfaced concrete failure modes: servo mounts lacked space for wires, a shared PWM channel made the elbow and base joints move in unison, and a 1,000 microfarad capacitor across the 5-volt and ground lines was needed to buffer current spikes from the servos.", "body_md": "## Meet Greg, the Robot With One Job\n\nMeet Greg, a robot with one singular purpose: detect a raised human palm and return a high-five. This simple, memorable interaction showcases advanced AI integration behind the scenes, delivering immediate, tangible results.\n\nGreg's architecture centers on five **servos** controlled by an **ESP32-S3** board. A secondary ESP32-CAM streams video over Wi-Fi to a Mac. This Mac runs Google MediaPipe, detecting open palms and calculating precise joint angles for the high-five.\n\nThis distributed system runs from a single power supply. Once powered on, Greg shifts into position, awaiting a human hand. The design demonstrates how AI can bridge mechanical design, firmware, and vision, but a human still built, tested, and assembled the machine.\n\n[Claude](https://www.stork.ai/en/claude-ai-workspaces) Opus 3.5 facilitated the entire design process, generating:\n\n- 3D models for parts\n- Wiring diagrams\n- Firmware for both ESP32 boards\n- Computer vision algorithms\n\nThe creator used an Elegoo Neptune 3 Pro 3D printer for manufacturing. Claude generated parts as Blender scripts, allowing direct manual adjustments. This project highlights a powerful workflow: AI-driven design combined with hands-on human assembly and iteration.\n\n## Claude Became the Robot’s Design Team\n\nClaude became Greg’s complete design team. Builder supplied concept and component dimensions; Claude generated all necessary artifacts. This included **Blender models**, wiring diagrams, firmware for both ESP32 boards, and vision code for the MediaPipe server.\n\nChoosing Blender was strategic. Builder avoided learning dedicated CAD software, instead leveraging familiar Blender. Claude generated designs via Python scripts, allowing direct inspection and manual adjustment if needed.\n\nPhysical feedback loops were critical. Before full-scale prints, builder printed small **servo-pocket tests**. Photos of fit problems went back to Claude, which analyzed images and revised designs. This iterative process saved hours of printing and filament on the Elegoo Neptune 3 Pro.\n\nThis workflow reduced design time and material waste. Claude accounted for precise component sizes like servos and the ESP32 camera, but sometimes missed details. For instance, an early shoulder design lacked space for servo wires; Claude corrected it after image feedback.\n\nThis multimodal iteration, feeding physical results back to the AI, proved essential. It demonstrated Claude’s capability to bridge digital design and physical prototyping, democratizing hardware development for complex projects like Greg.\n\n## The Tiny Mistakes That Stopped the Build\n\nPhysical build errors stopped progress. Claude’s initial servo mount designs omitted crucial space for wires, requiring a complete redesign that modeled the cable. Camera models also demanded real-world measurements after early 3D prints were too tight or too loose; Claude’s assumptions about thickness and lens position proved incorrect.\n\nElectrical challenges surfaced. Servos require a separate power supply path, never drawing directly from the ESP32’s 5-volt pin. All components must share a common ground; neglecting this caused trembling servos. A **1,000 microfarad capacitor** across the 5-volt and ground lines buffers sudden current spikes from multiple servos, preventing voltage dips that could reset the ESP32 and camera.\n\nFirmware presented a unique obstacle. The popular ESP32 servo library assigned both the elbow and base joints to the same **PWM channel**, causing them to move in unison. Explicitly assigning distinct PWM channels for each servo resolved this. Always verify AI-generated code; do not assume correct channel assignment, especially with shared libraries. For more details on vision system components, explore [MediaPipe Solutions - Google for Developers](https://developers.google.com/mediapipe).\n\nEnjoying this? Get one like it in your inbox each morning.\n\none email a day · unsubscribe in two clicks · no third-party tracking\n\n## From Awkward Arm to Working High-Five\n\nCalibrating Greg's arm required a grid-based approach. The arm sampled various positions while the camera tracked a piece of tape on its palm. This process mapped image coordinates directly to corresponding joint angles, establishing the foundational link between vision and motion.\n\nIterative refinements significantly improved Greg's high-five. A wider-angle camera enhanced detection, and voice commands allowed for precise real-time adjustments. Saved examples helped fine-tune wrist orientation, while adjustable sliders controlled delay and speed, personalizing the interaction.\n\nGreg ultimately achieved its high-fiving purpose, performing reliably after initial setup. However, it remained a **prototype**, occasionally exhibiting startup hiccups that required manual intervention.\n\nThis project highlights a crucial distinction: AI accelerates **creation**, generating designs and code rapidly. But the transformation from generated design to reliable hardware demands meticulous testing, precise measurements, and patient troubleshooting. Human oversight remains essential for converting AI-generated concepts into robust, functional physical systems.\n\n## Frequently Asked Questions\n\n### What does Greg the robot do?\n\nGreg detects a raised open palm and moves its five-servo arm to offer a high-five.\n\n### How does Greg recognize a raised hand?\n\nA camera streams video over Wi-Fi to a Mac, where Google MediaPipe detects the hand and helps calculate the arm’s target position.\n\n### What did Claude help create for the robot?\n\nClaude helped generate Blender models, wiring plans, firmware and computer-vision software, with the builder testing and correcting designs.\n\n### Can a beginner build an AI-designed robot?\n\nA motivated beginner with basic electronics experience can attempt it, but should expect hands-on assembly, careful power wiring and repeated fit tests.", "url": "https://wpnews.pro/news/claude-helped-build-a-robot-that-high-fives", "canonical_source": "https://www.stork.ai/blog/claude-helped-build-a-robot-that-high-fives", "published_at": "2026-11-05 15:22:16+00:00", "updated_at": "2026-11-05 16:23:38.735319+00:00", "lang": "en", "topics": ["artificial-intelligence", "robotics", "generative-ai", "computer-vision", "ai-tools"], "entities": ["Claude Opus 3.5", "Anthropic", "ESP32-S3", "ESP32-CAM", "Google MediaPipe", "Elegoo Neptune 3 Pro", "Blender"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/claude-helped-build-a-robot-that-high-fives", "markdown": "https://wpnews.pro/news/claude-helped-build-a-robot-that-high-fives.md", "text": "https://wpnews.pro/news/claude-helped-build-a-robot-that-high-fives.txt", "jsonld": "https://wpnews.pro/news/claude-helped-build-a-robot-that-high-fives.jsonld"}}