Project
AI Painter & Plotter
Electronics · 3D design · Python · RL
Overview
In this project, I successfully created an automated multi-tool gantry system capable of vectorizing digital images and physically rendering them using custom pen toolheads, while setting the mechanical foundation for airbrushing. The process involved developing a robust Python-based image processing pipeline to fine-tune key parameters, such as cross-hatching angles, stroke density, and simulated flow-field streamlines. Through these algorithmic adjustments, I was able to reliably translate complex digital images into optimized physical toolpaths that the robot executes on a physical canvas. One of the more challenging aspects was ensuring the seamless integration and precision of the custom hardware stack. I designed and built a backend control architecture that communicates with Klipper via a Unix Domain Socket (UDS) API, allowing me to orchestrate an STM32 board, air supply valves, and Arduino-driven servos with a high degree of accuracy, including down to 0.01 ml of volumetric control for the custom syringe pumps. This level of control was crucial for maintaining consistent fluid delivery and precise tool actuation. Additionally, I developed the complete software ecosystem to automate this process, using custom configuration loaders and automated G-code generation to bridge the gap between digital image processing and physical robotic motion. This project is currently ongoing. While the pen plotting mechanics are fully operational, I am currently iterating on the physical airbrush toolhead to achieve precise, real-time fluid and colour control. This ongoing hardware refinement is a critical roadblock to solve before deploying the AI component: a reinforcement learning pipeline designed to train an agent that optimizes painting strokes based on perceptual image quality. Once the physical fluid dynamics are fully stabilized, this RL policy will be able to step out of simulation and execute its digital twin paintings in the real world.
