This "Robot Vibe Coding" stream introduces "Tatbot," a personal project aimed at developing a robotic arm system for tattooing. The system utilizes Trossen Widow XAI robotic arms, an NVIDIA Jetson AGX Orin for edge computing, and a multi-camera setup (PoE IP and Intel RealSense) for real-time 3D reconstruction of the tattooing surface. The presenter discusses and demonstrates computer vision approaches for surface tracking, comparing models like "Master" and "VGGT," and explores using NVIDIA Warp's cloth simulation to project 2D tattoo designs onto 3D curved body parts. The stream also showcases the application of inverse kinematics (IK) via the Pyroki library for precise robot arm control. A key element of the presentation is a live "vibe coding" session using the Cursor AI assistant to debug Rust code for synchronized camera capture. The project emphasizes an open-source development philosophy, a self-funded, single-person business model focused on high-value artistic manipulation, and the strategic integration of emerging robotics foundation models for auxiliary tasks like wiping.
Tatbot: Developing an AI-Powered Tattoo Robot
0:03:57 Project Introduction: The stream, "Robot Vibe Coding," announces "Tatbot," a personal project to create a robot that tattoos. All project documentation and stream summaries are available on a public GitHub page (hoopoe/docs).
0:04:31 Tatbot Hardware & Setup: The system features 80/20 aluminum framing, two Trossen Widow XAI robotic arms, and an overhead multi-camera setup (PoE IP cameras for general viewing and Intel RealSense for depth). An NVIDIA Jetson AGX Orin (32GB unified RAM) serves as the edge computing device.
0:06:05 Tattooing Tools: The robot uses off-the-shelf tattoo wands/pens with external batteries and pre-made, sterile needle cartridges, leveraging existing industry tools rather than custom end effectors.
0:09:33 Distinguishing Tatbot from Existing Solutions: Current tattoo robots (e.g., 3D printer rigs, Black Dot's system) typically offer only 3D (XYZ) open-loop control, limiting them to flat surfaces. Tatbot aims for 6-degree-of-freedom (6D) control with its robotic arms, enabling tattooing on curved body parts.
0:12:57 Vision-Based Control for Precision: Due to the minuscule forces involved in tattooing, Tatbot relies on computer vision for closed-loop control rather than the limited force feedback capabilities of currently available robotic arms in its price range.
0:16:30 Real-Time 3D Skin Reconstruction: The project addresses the challenge of real-time skin tracking (at ~1Hz) to account for body deformation. Solutions explored include "Grounded Image Matching in 3D with Master" and "VGGT, Visual Geometry Grounded Transformers," both deep learning models for multi-view 3D reconstruction.
0:25:55 Custom Camera Capture Tool (R-Cam): A custom Rust-based tool, "R-Cam," was developed using AI-assisted "vibe coding" (with Cursor) to synchronously capture images from all IP and RealSense cameras. The presenter notes Rust's opinionated nature aids in generating higher-quality code through this workflow.
0:32:18 Live 3D Reconstruction Demo: A demonstration showcased 3D reconstructions from images captured by R-Cam using both VGGT (improved significantly after removing depth images) and Master, confirming their viability for surface tracking.
0:39:42 Projecting 2D Designs onto 3D Surfaces: The core challenge of transferring 2D tattoo designs onto complex 3D body shapes is being addressed using NVIDIA Warp's cloth simulation, which drapes a 2D design onto a 3D mesh to provide 6D pose information for each design point.
0:46:49 Inverse Kinematics (IK) for Tattooing Trajectories: Inverse kinematics is essential for calculating the precise joint angles required for the robot arm to reach each 6D point of the projected tattoo design. The Pyroki library (JAX-based) is demonstrated for smooth, GPU-accelerated 6D IK.
0:54:51 Live "Vibe Coding" Demonstration: The presenter performed a live "vibe coding" session, using Cursor AI to iteratively fix compiler warnings in the R-Cam Rust codebase, illustrating the back-and-forth interaction and the need to guide the AI assistant.
1:02:04 Strategic Use of Robotics Foundation Models: While precise tattooing relies on IK, robotics foundation models like NVIDIA Groot and Physical Intelligence Pi Zero are considered for less precise, "amorphous" tasks, such as a second robotic arm performing automated wiping of the tattooed area.
1:05:23 Project Philosophy: The project aims to be a self-funded, single-person endeavor to maintain creative control and avoid external pressures (e.g., from VCs) to expand into less personally appealing but potentially more profitable applications (like hair plugs).
1:27:49 Public Progress & Open Source: The first "embarrassing" Tatbot tattoo (a simple circle) was publicly shared on social media, aligning with a philosophy of early release and open-source development, encouraging others to copy and build upon the work.