Skip to content

Motion & Location Detection — Design Plan

How the Dilder detects movement, counts steps, senses gestures, and knows when you've gone somewhere new — all without GPS, using a $0.46 accelerometer and the ESP32-S3's built-in WiFi and BLE.

Table of Contents

  1. Design Goals
  2. Hardware — LIS2DH12TR Accelerometer
  3. Motion Detection — What the Accelerometer Provides
  4. Location Detection — WiFi Fingerprinting
  5. Location Detection — BLE Scanning
  6. Integration with Pet Logic
  7. Power Budget
  8. Why Not GPS?
  9. Implementation Phases
  10. Sources

1. Design Goals

The Dilder is a virtual pet. It doesn't need to know its precise latitude and longitude. It needs to answer simpler questions:

Question Method Hardware
Is the user walking? Step counter (hardware pedometer) LIS2DH12
Did the user shake me? Tap/shake interrupt LIS2DH12
Am I being carried or sitting still? Activity/inactivity detection LIS2DH12
Am I at the same place as before? WiFi AP fingerprint delta ESP32-S3 WiFi (built-in)
Are other Dilders nearby? BLE scan for peer devices ESP32-S3 BLE (built-in)
Am I upside down? Orientation detection LIS2DH12

All of this is achieved with zero additional hardware beyond the LIS2DH12 — the ESP32-S3 provides WiFi and BLE natively.

2. Hardware — LIS2DH12TR Accelerometer

Why LIS2DH12TR

The original design used the MPU-6050 (6-axis IMU with gyroscope) at $6.88. The gyroscope is unnecessary for step counting and gesture detection. The LIS2DH12TR is a 3-axis accelerometer with a built-in hardware pedometer that handles step counting in silicon, waking the MCU only when needed.

Specifications

Attribute Value
Part STMicroelectronics LIS2DH12TR
Axes 3 (X, Y, Z accelerometer only — no gyroscope)
Resolution 12-bit (high-resolution mode), 10-bit, or 8-bit selectable
Range +/-2g, +/-4g, +/-8g, +/-16g selectable
Interface I2C (up to 400kHz) / SPI (up to 10MHz)
Address 0x18 (SA0=GND) or 0x19 (SA0=VCC)
Package LGA-12 (2x2mm) — same footprint as MPU-6050's QFN-24 is larger, this is smaller
Supply 1.71V to 3.6V
Current (normal) 185uA @ 400Hz ODR
Current (low power) 2uA @ 1Hz ODR
Current (power down) 0.5uA
Built-in features Pedometer (step counter + step detector interrupt), click/double-click detection, free-fall detection, wake-up/activity recognition, 6D/4D orientation detection, temperature sensor
FIFO 32-level, reduces I2C/SPI traffic
LCSC C110926
Price $0.46 (qty 5)

Pin Connections (Dilder Board)

LIS2DH12 Pin ESP32-S3 GPIO Function
SDA/SDI GPIO16 I2C data
SCL/SPC GPIO17 I2C clock
SA0/SDO GND Sets I2C address to 0x18
CS VCC (3.3V) Pulled high to select I2C mode (not SPI)
INT1 GPIO18 (TBD) Step detector interrupt / activity wakeup
INT2 N/C Not connected (single interrupt sufficient)
VDD 3.3V Power
GND GND Ground

Cost Savings

Component Old New Savings
Sensor IC MPU-6050 ($6.88) LIS2DH12TR ($0.46) -$6.42
I2C pull-ups 2x 10k (same) 2x 10k (same) $0.00
GPS module ATGM336H ($1.80 planned) Dropped -$1.80
Total sensor savings -$8.22/board

3. Motion Detection — What the Accelerometer Provides

3.1 Step Counting (Hardware Pedometer)

The LIS2DH12 has a built-in step counter that runs in hardware even when the ESP32-S3 is in deep sleep:

  • The pedometer algorithm runs inside the LIS2DH12 at ~2uA
  • A step counter register (16-bit) accumulates steps
  • INT1 can fire a step detector interrupt on each detected step, or a significant motion interrupt after a configurable threshold
  • The ESP32-S3 reads the step count register during each wake cycle (e.g. every 10 minutes in Tamagotchi mode)

No software step-counting algorithm needed. The LIS2DH12 handles it in silicon.

ESP32-S3 (deep sleep, 10uA)     LIS2DH12 (pedometer active, 2uA)
         |                                |
         |     [user walks 500 steps]     |
         |                                | step_count += 500
         |                                |
    [wake every 10 min] ────── I2C read ──► step_count = 500
         |                                |
    "You walked 500 steps!"               |
    [update display, sleep]               |

3.2 Shake / Tap Detection

The LIS2DH12 has hardware click/double-click detection on all three axes:

  • Single tap — user taps the device (e.g. "interact with the octopus")
  • Double tap — user double-taps (e.g. "feed the octopus")
  • Shake — high-g activity on multiple axes (e.g. "wake up!" or "you're scaring me")

Each generates a hardware interrupt — the ESP32-S3 doesn't need to poll.

3.3 Activity / Inactivity Detection

  • Activity interrupt — fires when acceleration exceeds a threshold (user picked up the device)
  • Inactivity interrupt — fires when acceleration stays below threshold for N seconds (device put down)

This lets the pet know: - "I'm being carried" (activity + steps) - "I'm sitting on a desk" (inactivity) - "I'm in a pocket" (occasional movement but no steps)

3.4 Orientation Detection (6D)

The LIS2DH12 can detect which face of the device is pointing up:

  • Face up — display visible (normal operation)
  • Face down — display hidden (sleep mode? privacy mode?)
  • Tilted left/right/forward/back — could map to pet interactions

3.5 Free-Fall Detection

Detects when the device is in free fall (all axes read ~0g). Could trigger: - "You dropped me!" reaction - Save state before potential impact

4. Location Detection — WiFi Fingerprinting

How It Works

The ESP32-S3 scans for nearby WiFi access points and records their BSSIDs (MAC addresses) and signal strengths (RSSI). This creates a "fingerprint" of the current location.

Location A (home):          Location B (office):
  BSSID_1: -45 dBm           BSSID_5: -30 dBm
  BSSID_2: -60 dBm           BSSID_6: -55 dBm
  BSSID_3: -70 dBm           BSSID_7: -70 dBm

  Hash: 0xA3F2                Hash: 0x7C91
  → Different hashes = different location!

Algorithm

  1. During each active wake cycle, scan WiFi (takes ~2 seconds)
  2. Collect top-5 strongest BSSIDs + their RSSI values
  3. Hash the BSSID list → produces a location fingerprint
  4. Compare with the previous fingerprint:
  5. Same hash → same location → pet is stationary
  6. Different hash → new location → pet was taken somewhere!
  7. No APs found → outdoors or rural → pet is on an adventure

Accuracy

Environment Accuracy Notes
Dense urban ~20-50m Many overlapping APs
Suburban ~50-100m Fewer APs, wider spacing
Rural / outdoors Cannot detect No WiFi APs to fingerprint
Indoors (same building) Room-level Different rooms have distinct RSSI patterns

Power Cost

  • WiFi scan: ~80mA for ~2 seconds = ~0.044 mAh per scan
  • At 1 scan per 30 minutes: ~0.088 mAh/hour — negligible impact on battery life
  • Compare to GPS: 25mA continuous tracking = orders of magnitude more power

What It Can't Do

  • No absolute coordinates (lat/lon) — just "same place" vs "different place"
  • Doesn't work in areas with no WiFi coverage
  • Can't measure distance traveled

For absolute coordinates, use the WiFi Geolocation API (Option 3 below) which sends the BSSID list to a cloud service and gets back approximate lat/lon. This requires an internet connection but is free (Google: 40k requests/month, Mozilla Location Service: unlimited).

5. Location Detection — BLE Scanning

How It Works

The ESP32-S3 scans for nearby BLE advertisements. This serves two purposes:

  1. Peer discovery — Find other Dilder devices for social interactions (already planned for the engagement system)
  2. Environment fingerprinting — BLE devices (phones, smart watches, fitness trackers) create a rough signature of the current environment

Power Cost

  • BLE scan: ~15mA for ~3 seconds = ~0.012 mAh per scan
  • ~5x more power-efficient than WiFi scanning

Best Used For

  • Detecting other Dilders nearby (mating system, social interactions)
  • Counting nearby BLE devices as a proxy for "busy location" vs "quiet location"
  • Augmenting WiFi fingerprinting with BLE data for better location delta detection

6. Integration with Pet Logic

Sensor Events → Pet Behaviors

Sensor Event Pet Interpretation Game Effect
Steps > 500 in a wake cycle "We went for a walk!" +happiness, +health, unlock "adventure" dialogue
Steps > 2000 in a day "Big adventure!" +XP, possible evolution trigger, treasure hunt progress
Shake detected "Hey! That's rude!" or "Wheee!" Mood-dependent reaction
Single tap "Oh, a pet!" +happiness (if not angry)
Double tap "Feed me!" Trigger feeding interaction
Free fall "AAAAAHHH!" -happiness, panic expression
Face down "It's dark in here..." Enter sleep/sulk mode
WiFi fingerprint changed "New place! What's this?" +curiosity, new location dialogue, exploration XP
WiFi fingerprint same (24h+) "We never go anywhere..." -happiness (homesick/bored)
BLE: other Dilder found "A friend!" Trigger peer interaction / mating
Inactivity > 1 hour "...hello? Anyone there?" Lonely mood, attention-seeking quotes

Data Persistence

Store in ESP32-S3 NVS (non-volatile storage): - daily_steps — reset at midnight - total_steps — lifetime counter - last_wifi_hash — for location delta comparison - locations_visited — count of unique WiFi fingerprints seen - last_activity_time — for inactivity detection

7. Power Budget

Motion + Location Detection Power

Source Current Duration Frequency Avg mA
LIS2DH12 pedometer (always on) 2uA Continuous Always 0.002
LIS2DH12 interrupt wake 0uA (ESP32 handles) — Per-step 0.000
WiFi scan (location fingerprint) 80mA 2 sec Every 30 min 0.089
BLE scan (peer discovery) 15mA 3 sec Every 30 min 0.025
I2C step count read 5mA 10ms Every 10 min 0.000
Total detection overhead ~0.116 mA

This adds ~0.1 mA average to the power budget — less than 2% of the Tamagotchi mode baseline (~5.5 mA average). Battery life impact: effectively zero.

Comparison with GPS

Method Avg current Battery impact (1000mAh)
LIS2DH12 + WiFi/BLE scans ~0.1 mA Negligible (<2%)
GPS (always tracking) 25 mA -4.3 days battery life
GPS (periodic fix, 1/30min) ~1.5 mA avg -1.1 days battery life

8. Why Not GPS?

Factor GPS WiFi + Accelerometer
Component cost $1.80-4.00 $0.00 (WiFi built-in)
Power draw 20-45mA active 0.1mA average
Board space ~12x12mm module + antenna 0mm (no additional component)
Works indoors No (needs sky view) Yes
Cold start time 30-60 seconds Instant
Accuracy outdoors 2.5m ~50-100m (WiFi), or step count only
Accuracy indoors Doesn't work Room-level (WiFi fingerprint)
Complexity UART driver + NMEA parsing + antenna design WiFi scan API (built-in)

GPS is overkill for a virtual pet. The pet doesn't need to know it's at 48.1351°N, 11.5820°E. It needs to know "you took me somewhere new" and "you're walking." WiFi fingerprinting + step counting provides exactly that at zero additional cost and negligible power.

GPS stays in the design as a Phase 7 optional add-on for users who want real geocaching/treasure-hunt features. The board can be designed with unpopulated GPS pads for future hand-soldering.

9. Implementation Phases

Phase 2 (Current) — Hardware Foundation

  • [x] LIS2DH12TR selected and added to BOM ($0.46 replaces $6.88 MPU-6050)
  • [x] I2C wiring planned (GPIO16/17, same bus as the old MPU-6050 plan)
  • [ ] Update KiCad schematic with LIS2DH12 footprint and connections
  • [ ] Verify LIS2DH12 LCSC stock and JLCPCB assembly compatibility

Phase 3 — Pet Logic (Firmware)

  • [ ] Initialize LIS2DH12 over I2C (set ODR, range, enable pedometer)
  • [ ] Enable step detector interrupt on INT1
  • [ ] Read step count register during each wake cycle
  • [ ] Implement shake/tap detection via click interrupt
  • [ ] Map accelerometer events to pet mood changes

Phase 4 — Location Awareness (Firmware)

  • [ ] Implement WiFi AP scan during active wake
  • [ ] Hash top-5 BSSIDs into location fingerprint
  • [ ] Store fingerprint in NVS, compare with previous
  • [ ] Trigger "new location" pet event on fingerprint delta
  • [ ] Implement BLE scan for peer Dilder discovery

Phase 7 — Optional GPS Expansion

  • [ ] Design unpopulated GPS pads on PCB (ATGM336H footprint)
  • [ ] Implement UART GPS driver (for users who hand-solder the module)
  • [ ] WiFi Geolocation API integration (cloud-based lat/lon from BSSID list)
  • [ ] Geocaching / treasure hunt features using approximate coordinates

10. Sources

Document version: 1.0 — 2026-04-15