Autotune: Vehicle-Specific Threshold Calibration

[jctoledo]

Autotune Feature - Per-Mode Calibration

This enables users to tune the parameters for event detection (ACCEL, BRAKE, CORNER, etc) to their specific vehicle. Simply go to /autotune and start calibration by driving around. Settings get saved to NVS and are ready to use on /

Changes

• /autotune page with real-time event detection UI
• Independent NVS storage for each driving mode (Track, Canyon, City, Highway, Custom)
• Active mode persists across power cycles
• Session persistence during calibration (sessionStorage, 1-hour expiry)
• Removed dead code: PROFILE_SCALES, generateAllProfiles(), MIN_SPEED_MS
• Synchronized defaults between settings.rs and dashboard JS
• Added From trait impls to reduce type conversion boilerplate

Autotune UI (/autotune)

Workflow:

1. Select mode to calibrate (City/Canyon/Track/Highway)
2. Start calibration and drive in that style for 10-20 minutes
3. UI detects and counts accel/brake/turn events in real-time
4. Live display shows EMA-filtered longitudinal G, lateral G, centripetal acceleration, speed
5. Progress bar and confidence indicator (LOW/MEDIUM/HIGH based on event counts)
6. Apply saves thresholds to ESP32 NVS + browser localStorage

Event Detection (matches mode.rs dynamics):

• EMA α=0.85 tuned for 25Hz polling to simulate 200Hz firmware behavior
• Noise floor 0.08g rejects road bumps
• 200ms minimum duration, 400ms cooldown between events
• Turn detection requires lateral G + yaw rate sign consistency

Threshold Calculation:

• IQR outlier rejection on collected peak values
• Entry = median × 0.7
• Exit = entry × 0.5 (hysteresis)

Session Persistence: Navigate away and back within 1 hour - calibration progress restored (events, selected mode, recording state).

Analysis

Value proposition: Vehicle/driver-specific threshold tuning. A sports car pulling 1.0g lateral needs different thresholds than a minivan at 0.5g.

Target users: Track/autocross enthusiasts analyzing lap data. For casual users, default presets work fine.

Future potential: Event detection infrastructure supports more granular labels:

• Heavy/moderate/light braking intensity
• Corner entry vs apex vs exit phases
• Trail braking detection (brake + turn overlap)
• Lift-off oversteer warning (decel + high yaw rate)

Peak values, duration, and delta-V are already captured per event.

Test Sequence

espflash erase-flash && cargo espflash flash --release --monitor

1. Dashboard (/) → tap "Tune" → opens /autotune
2. Select Track, tap "Start Calibration", drive aggressively
3. Watch event counters increment, progress bar fill
4. Tap "Apply" → confirms save, returns to dashboard
5. Dashboard → tap City preset → independent City values load
6. Tap "Tune" → select City, calibrate with normal driving, Apply
7. Power cycle → last active mode and its calibrated settings restored
8. Dashboard preset buttons load each mode's saved thresholds