AirMouse-3D: An On-Device TinyML Inertial Mouse for Table-Free Desktop Interaction

Humans prefer unconstrained, free-space movement—so why must the mouse stay on a tabletop? This paper presents the design and development of a novel three-dimensional (3D) motion-based mouse that operates without a surface, built around the Arduino Nano 33 BLE Sense and Google’s Tiny Motion Trainer. The system uses on-board inertial sensing to capture roll, pitch, yaw, and small lateral/vertical translations, and employs TinyML classification to map these motions to discrete desktop actions. Motion-command map used in this study: pitch↑ → scroll up; pitch↓ → scroll down; roll→ → left-click; roll← → right-click; yaw→/yaw← → drag toggle on/off; lateral± → cursor nudge ±Δx; vertical± → cursor nudge ±Δy. The device is housed in a 3D-printed hexagonal-prism casing with ergonomic circular cuts for stable grip and repeatable gestures, and includes an LED and buzzer for immediate user feedback. The development pipeline comprised (i) gyroscope/IMU calibration and real-time motion mirroring in Processing, (ii) enclosure design and 3D printing, (iii) gesture dataset collection and model training in Tiny Motion Trainer, and (iv) Python integration over serial (pyserial) to synthesize OS-level inputs (pynput). Compared to conventional mice, the proposed interface enables multi-dimensional, touch-free interaction from sofas, beds, or standing postures, removing surface constraints while preserving familiar desktop actions. We detail the hardware, firmware, and TinyML workflow, discuss practical considerations (drift, debouncing, gesture separability, and comfort), and outline evaluation protocols and extensions (adaptive thresholds, continuous cursor control, and user-specific calibration) to advance free-motion pointing.