Build a plant-biosignal sensor with M5Stack and ESPHomeBuild a plant-biosignal sensor with M5Stack and ESPHome

Purpose and scope

Plants generate tiny electrical signals. Ions move across cell membranes when the plant responds to light, water, wounding or nutrient change, and that movement shows up as a sub-millivolt voltage you can read with electrodes on the stem^[3]. Commercial units like the Vivent VITA1 do exactly this, then use trained models to infer water and stress state from how the signal drifts^[5].

Electrically, reading a plant is the same problem as reading a heartbeat: a small, noisy, high-impedance voltage you must amplify cleanly. That means the cheap, proven ECG front-end chip, the AD8232, works straight out of the box for plants^[1]. Bolt it to an ESP32 and ESPHome and you have a logging plant-biosignal sensor for about NZ$110.

Definitions

Evidence and limitations

The signal chain

Everything is one line from plant to dashboard. Two electrodes sense the stem, one references the soil; the AD8232 amplifies and filters; the ADS1115 digitises; the ESP32 publishes.

Bill of materials

About NZ$110. M5Stack parts come as plug-together Grove modules; the two non-M5Stack parts (the AD8232 front-end and the ADS1115 ADC) are the standard cheap substitutes, because nobody sells a biopotential amplifier in the Grove ecosystem. Prices are indicative, mid-2026.

Wiring

Everything downstream of the controller is one I2C bus, fanned out by the HUB. The single analog wire in the whole build is the AD8232 output into the ADS1115's A0 pin. Electrodes plug into the AD8232's three-pin header.

Electrodes and placement

Contact impedance is the resistance between the electrode metal and the plant's tissue. Think of it like a loose headphone jack — a poor connection weakens the signal and lets noise in before it reaches the amplifier. The AD8232 handles everything downstream; your electrode contact quality is the one variable you control. Two workable options^[4]:

1. 1
   Place the pair along the stem

   LA (sense +) on the upper stem near a node; RA (sense −) 5–15 cm (2–6 in) lower on the same stem. This pair captures the travelling signal.
2. 2
   Reference into the soil

   RL (the driven reference) goes into the moist root-zone soil. It is what cancels common-mode hum, so don't skip it.
3. 3
   Gel and tape

   A dab of Ten20 under each surface contact, then micropore tape for light, steady pressure. A rising, noisy baseline is usually a drying electrode, not a sick plant.
4. 4
   Let it settle

   When metal contacts wet tissue, ions exchange at the surface and build a small voltage of their own — a half-cell potential, the same effect that makes a lemon and two different coins into a battery. Allow 10–30 minutes for this to stabilise. The initial drift is electrode chemistry settling, not plant activity.

ESPHome firmware

The full config. Put your Wi-Fi, API and OTA secrets in ESPHome's secrets.yaml, flash from the ESPHome dashboard, and it auto-discovers into Home Assistant. On a StampS3 or Core2, change the board line and, if needed, the two I2C pins.

esphome:
  name: plant-biosignal
  friendly_name: Plant Biosignal

esp32:
  board: m5stick-c            # Core2 -> m5stack-core2 ; Stamp S3 -> m5stack-stamps3
  framework: { type: arduino }

logger:
api:
  encryption: { key: !secret api_key }
ota:
  - platform: esphome
    password: !secret ota_pw
wifi:
  ssid: !secret wifi_ssid
  password: !secret wifi_password

# ---- I2C bus (Grove Port A on StickC PLUS2 / Core2) ----
i2c:
  sda: 32
  scl: 33
  frequency: 400kHz
  scan: true

# ---- 16-bit ADC hub ----
ads1115:
  - address: 0x48

sensor:
  # === PLANT BIOPOTENTIAL ===
  - platform: ads1115
    multiplexer: 'A0_GND'       # AD8232 OUTPUT on A0, single-ended
    gain: 4.096                 # FSR covers the 0-3.3V swing
    name: "Plant Biopotential (raw)"
    id: bio_raw
    unit_of_measurement: "V"
    accuracy_decimals: 4
    update_interval: 1s
    filters:
      - median: { window_size: 5, send_every: 1 }   # kill spikes

  # Centre on the ~1.5V bias -> signed millivolts (the number you watch)
  - platform: template
    name: "Plant Biopotential"
    id: bio_mv
    unit_of_measurement: "mV"
    accuracy_decimals: 2
    update_interval: 1s
    lambda: |-
      return (id(bio_raw).state - 1.5) * 1000.0;   // trim 1.5 to your measured baseline

  # Slow envelope = the "trend" a commercial unit charts every 5 min
  - platform: template
    name: "Plant Signal Trend"
    unit_of_measurement: "mV"
    accuracy_decimals: 2
    lambda: |- return id(bio_mv).state;
    filters:
      - exponential_moving_average: { alpha: 0.02, send_every: 30 }

  # === ENVIRONMENT (ENV IV Unit) ===
  - platform: sht4x
    address: 0x44
    temperature: { name: "Grow Air Temp", id: air_t }
    humidity:    { name: "Grow Humidity", id: air_rh }
  - platform: bmp280_i2c        # older ESPHome: use "bmp280"
    address: 0x76
    temperature: { name: "Grow Baro Temp" }
    pressure:    { name: "Grow Pressure" }

  # === LIGHT (DLight Unit) ===
  - platform: bh1750
    address: 0x23
    name: "Grow Light"
    update_interval: 10s

  # === DERIVED: VPD (kPa) ===
  - platform: template
    name: "Grow VPD"
    unit_of_measurement: "kPa"
    accuracy_decimals: 2
    update_interval: 10s
    lambda: |-
      float t = id(air_t).state;
      float rh = id(air_rh).state;
      float es = 0.6108 * exp(17.27 * t / (t + 237.3));  // saturation vapour pressure
      return es * (1.0 - rh / 100.0);

# === OPTIONAL: electrode lead-off detection ===
binary_sensor:
  - platform: gpio
    pin: 26                       # AD8232 LO+ ; pick a free GPIO for your board
    name: "Electrode Lead-off"
    device_class: problem

Interpreting plant-biosignal data

These are changes in potential, not a calibrated physiological unit. Compare a plant to itself over time, never plant-to-plant in raw millivolts.

To approach a commercial unit's inference, log the raw trace alongside VPD, light and irrigation events for a few weeks, then hunt for your own repeatable correlations, for example signal amplitude collapsing before visible wilt as an early water-stress warning. That correlation library is exactly what the paid product ships pre-built. Feed the trace into the plant-state dashboard to combine it with your other telemetry.

Limitations, calibration and safety