A smart pen that writes on real paper with a standard gel pen refill — while streaming real-time sensor data wirelessly to a host computer, where a deep learning model reconstructs your handwriting as digital strokes on any canvas application.
The pen registers as an OS-level input device (like a Wacom pen), so it works with any drawing or writing app out of the box.
Real Paper Smart Pen Wireless Host Computer
┌──────────┐ ┌────────────────┐ ┌──────────────┐ ┌────────────────┐
│ Write │───▶│ 6 sensors │───▶│ 2.4GHz ESB │───▶│ ML inference │
│ normally │ │ stream at 8kHz │ │ at 2Mbps │ │ < 10ms latency │
└──────────┘ └────────────────┘ └──────────────┘ └────────────────┘
│
▼
┌────────────────┐
│ OS input device│
│ (virtual pen) │
│ ─── any app ───│
└────────────────┘
No special paper. No camera. No digitizer tablet. Just pick up the pen and write.
The entire electronics package fits inside a pen that looks and feels like a premium writing instrument.
| Parameter | Value |
|---|---|
| Dimensions | 150mm x 11.5mm (similar to Lamy Safari) |
| Weight | 20-28g |
| Refill | Standard gel pen (Pilot G2 compatible) |
| Sensors | 2x IMU (32kHz), pressure, capacitive touch |
| MCU | nRF52840 (ARM Cortex-M4F, integrated 2.4GHz) |
| Wireless | Nordic ESB at 2Mbps, BLE secondary |
| Streaming | 8kHz sensor polling, 6 channels |
| Battery | 10180 Li-ion (100mAh), 8-14 hours active |
| Charging | Qi wireless, ~30 min charge time |
| Shell | Polycarbonate or Aluminum 6061 |
The pen houses a helical flex PCB (polyimide, 230mm x 10mm) that wraps around an inner cylindrical shell, placing components in 6 functional zones along the pen axis:
- Zone A (nib end): Pressure sensor + IMU #1
- Zone B (grip area): Capacitive touch slider + wake controller
- Zone C (mid-body): nRF52840 MCU + 2.4GHz antenna
- Zone D (upper body): Power management + IMU #2
- Zone E (end cap): Qi wireless charging coil
Phase: Research Complete → Design Phase Starting
The project has a comprehensive knowledge base with 50+ research documents covering four domains. No code or firmware exists yet — the next step is schematic capture and mechanical CAD.
| Domain | Status | Key Decision |
|---|---|---|
| Hardware | Research complete | Flex PCB + ICM-42688-P IMU + nRF52840 |
| Embedded | Research complete | Zephyr RTOS + ESB wireless + DMA-driven ADC |
| ML | Research complete | MoE architecture (TCN + RT-TCN streaming) |
| Software | Research complete | HID Usage Page 0x0D + virtual pen per OS |
This repository is an Obsidian vault — open the docs/ folder in Obsidian for the best navigation experience.
docs/
├── Home.md # Vault entry point
├── 01-Project/ # Vision, requirements, architecture
│ ├── 3d-pen-MOC.md # Master Map of Content (start here)
│ ├── vision.md # Product concept and constraints
│ ├── requirements.md # Functional & non-functional requirements
│ ├── architecture.md # System design and data flow
│ └── hardware-design-spec.md # Full hardware geometry for CAD/renders
├── 02-Research/ # Domain research (50+ notes)
│ ├── hardware/ # Flex PCB, sensors, charging, mechanical
│ ├── embedded/ # MCU, wireless, firmware, power
│ ├── ml/ # Sensor fusion, training, inference
│ └── software/ # HID, drivers, canvas, OS integration
├── 03-SOPs/ # 10+ standard operating procedures
├── 04-Tools/ # Tool evaluations
├── 05-Plans/ # Phase plans
├── 06-Decisions/ # Architecture decision records
├── 07-References/ # Papers, datasheets
├── 08-Templates/ # Note templates
└── 09-Assets/images/ # Sketches, renders
├── ai-renders/ # AI-generated concept renders
├── refil/ # Reference photos of gel pen refills
└── pen-sketch-v0.jpg # Original hand-drawn sketch
| Document | Description |
|---|---|
| Hardware Design Spec | Complete dimensions, materials, cross-sections for CAD modeling |
| Architecture | System block diagram, data flow, interface definitions |
| Requirements | 14 functional + 9 non-functional requirements |
| Sensor Selection | ICM-42688-P, FSR 400, IQS263 evaluation |
| MCU Selection | nRF52840 vs alternatives analysis |
| Sensor Fusion Models | MoE architecture, TCN displacement mapping |
| HID Protocol | USB HID digitizer spec for pen input |
# Clone the repository
git clone https://github.com/Bunty9/3d-pen.git
# Open docs/ as an Obsidian vault
# 1. Install Obsidian: https://obsidian.md/
# 2. Open Vault → select the docs/ folder
# 3. Navigate from Home.mdSee CONTRIBUTING.md for guidelines on contributing to any domain.
See Developer Documentation for setup, vault conventions, and agent workflows.
This project uses a multi-agent workflow where specialized AI agents work on different domains with exclusive write scopes. See SOP-0002: Multi-Agent Orchestration for the full protocol.
| Agent | Domain | Write Scope |
|---|---|---|
| hardware-agent | PCB, sensors, mechanical | 02-Research/hardware/ |
| embedded-agent | MCU, firmware, wireless | 02-Research/embedded/ |
| ml-agent | Models, training, inference | 02-Research/ml/ |
| software-agent | HID, drivers, canvas | 02-Research/software/ |
| integration-agent | Cross-domain alignment | 01-Project/architecture.md |
- Phase 1: Research — Knowledge base with 50+ documents
- Phase 2: Design — Schematic capture, mechanical CAD, firmware architecture, ML model selection
- Phase 3: Prototype — PCB fabrication, component sourcing, firmware MVP, training data collection
- Phase 4: Integration — End-to-end sensor-to-canvas pipeline working
This project is currently in early development. License TBD.
Research informed by analysis of STABILO DigiPen, D-POINT, Livescribe, Neo Smartpen, and the OnHW dataset (Fraunhofer IIS / ACM IMWUT 2020). AI renders generated with Gemini and ChatGPT.