GeardUp!

Inspiration

Many people with limited hand or arm mobility—due to spinal injury, ALS, stroke, or other conditions—miss out on experiences the rest of us take for granted, including driving and even driving simulations. We wanted to give them a way to control a car in a game. GeardUp! started from the question: Can we use affordable biosignal hardware (EMG from a SpikerBox) to simulate driving and to communicate—without needing fine motor control or voice?

What it does

GeardUp! is a biosignal-based control and communication suite:

Keyboard control (driving simulation) — Two EMG channels from a Human SpikerBox are streamed in real time, bandpass-filtered, and mapped to keyboard keys: Channel 0 → 'd' (e.g. steer right), Channel 1 → 'a' (steer left), both active → 'w' (e.g. accelerate). This is aimed at simulating driving for disabled users: they can play driving games or use driving sims by flexing different muscles instead of using a keyboard or wheel. A short calibration (hold still for a few seconds) sets a per-channel baseline; an adaptive baseline and hysteresis reduce false triggers and keep control stable.

How we built it

Hardware: Human SpikerBox for 2-channel EMG (and optional EEG) at 5 kHz; Arduino Uno Q for motor/vehicle control.
Keyboard control pipeline: Python streams EMG from the SpikerBox over serial, applies a 50–400 Hz bandpass, and computes a level metric (we use p95 of the filtered signal) per chunk. We use percentile-based rest detection (e.g. “at rest” when level ≤ 75th percentile of a sliding window) and update the baseline with an EMA only when at rest. Activation uses a configurable threshold above baseline with hysteresis (separate deactivate threshold) so keys don’t flicker. We simulate key presses with pynput (Ch0 → d, Ch1 → a, both → w), so any driving game or sim that uses these keys works without modification.
Calibration: A separate step records a few seconds of rest-only EMG to compute per-channel baseline and noise standard deviation; this is saved and loaded at runtime for robust activation thresholds.

Challenges we ran into

Baseline drift — EMG baseline changes with posture and fatigue. We addressed this with an adaptive baseline (EMA updated only when “at rest”) and optional calibration so the system stays usable over time.
Rest detection — Deciding when the user is “at rest” without a separate rest gesture was tricky. We used a sliding window of recent levels and a percentile cutoff so we don’t need a fixed absolute threshold.
False activations — Single spikes or noise could trigger keys. Using the p95 level metric instead of max, plus hysteresis (higher threshold to activate, lower to deactivate), and noise-based threshold scaling from calibration all helped.
Mapping both channels — When both muscles fire (e.g. “go forward”), we map to a single key (w) instead of pressing both a and d, so the game receives a clear “accelerate” intent.

Accomplishments that we're proud of

Real-time EMG-to-keyboard for driving sims with no game-specific integration—any title that uses WASD works.
Calibration + adaptive baseline so the same setup works across users and sessions without constant retuning.
Clear use case for simulating driving for disabled people using only two EMG channels and a SpikerBox.
Unified project that connects keyboard control (sims), silent speech (communication), and Arduino Q (hardware control) under one accessibility-focused vision.

What we learned

Rest detection via percentiles over a short history works well for EMG when you don’t have a dedicated “rest” gesture.
Hysteresis and a robust level metric (p95) are essential for stable, flicker-free key output.
Affordable hardware (SpikerBox) is enough to get usable, low-latency control for driving sims and communication with the right signal processing.

What's next for GeardUp!

Tighter driving-sim integration: Optional overlay or plugin for popular sims (e.g. steering wheel API) so EMG can map to analog steering and throttle instead of discrete keys.
More channels and gestures: Support for more EMG channels and distinct gestures (e.g. brake, handbrake) for richer driving control.
User studies: Partner with accessibility groups to test keyboard control and silent speech with people who have limited mobility or speech.
Arduino Q in the loop: Use the same EMG processing ideas from keyboard control in the Arduino Uno Q runtime for real steering and throttle in a safe, controlled environment (e.g. go-kart or simulator rig).