ESP32 PWM motor control + Python host system with Flask UI, vision tracking, ML perception, and CI/CD

[Copilot]

Upgrades the ESP32 firmware from binary digitalWrite motor control to PWM via ledcAttach/ledcWrite, and builds out the Python host system: weighted face tracking, Flask control panel, optional ML perception (MediaPipe/DeepFace), and GitHub Actions CI.

Firmware (sylvie_main.ino)

• PWM motor control: setMotor(motor, dir, speed) using ledcWrite with 8-bit resolution (0–255), replacing digitalWrite on/off
• OSC speed parameter: /motor1 and /motor2 now accept [dir, speed] (speed defaults to 255 for backward compat)

// Before: full speed or stop only
void setMotor(int motor, int direction) {
  digitalWrite(pinA, direction > 0 ? HIGH : LOW);
}

// After: PWM speed control via L298N
void setMotor(int motor, int direction, int speed = 255) {
  if (direction > 0) { ledcWrite(pinA, speed); ledcWrite(pinB, 0); }
}

Python Host (python_host/)

• vision/face_tracker.py — Weighted multi-face selection: weight = area × 1/(1 + dist/max_dist). Thread-safe JPEG streaming. Multi-camera hot-switching (iPhone Continuity Camera support).
• vision/perception.py — Lazy-loaded MediaPipe face mesh/pose + DeepFace emotion. Graceful degradation when ML deps absent. Separate requirements-ml.txt to avoid dependency hell.
• network/osc_sender.py — Thread-safe OSC sender with override flag (manual UI blocks auto-tracking commands). stop_all() bypasses override. TFT eye animation stub reserved.
• ui/app.py + index.html — Flask control panel: live video feed, motor dir+speed sliders, 2D XY pad (speed/jitter), LED color picker, presets, Override toggle, Tag & Save (JSONL data labeling for future training).

CI/CD (.github/workflows/python-ci.yml)

• Core tests on Python 3.10/3.11/3.12 (no ML deps)
• ML integration tests on 3.11
• flake8 lint
• 27 tests covering weight algorithm, OSC sender override logic, Flask API endpoints, perception graceful degradation

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.