Back to all projects
Grove Vision AI starter kit with ESP32-C3 SuperMini, Seeed Grove Vision AI Module V2 camera, and OLED display

Intermediate

Build a Grove Vision AI Starter

SensorsIoT

Add a dedicated vision brain to any controller: a co-processor with its own onboard NPU that runs real inference on its own, and just hands the controller a finished answer instead of raw pixels to think about.

2 real parts included

Total for this kit: 1,784.00 ฿

Doesn't include any order discount you may get at checkout.

What you'll need

Project Goal

This project pairs a general-purpose controller with a dedicated vision co-processor, a module that has its own onboard NPU and does its own thinking. The controller doesn't process any image data itself, it just asks the vision module a question and gets a real answer back, the same division of labor real production vision systems use.

What the Controller Does

It periodically asks the vision module for its latest result over a simple interface, then acts on whatever answer comes back, lighting an LED, printing to a display, or triggering any other output already in the parts bin.

What the Vision Module Does

It captures its own images and runs a real inference model against them entirely on its own onboard NPU, independent of whatever is asking it questions. This is a genuinely different architecture from a camera board wired directly to a controller: the heavy computational work never touches the host at all.

Software and Library Concepts

The module ships with pre-trained models ready to run immediately, giving a working result on day one, with the option to train and deploy a custom model later once the basic communication pattern with the controller is proven out.

Testing and Troubleshooting

  • No response from the module, confirm wiring and the correct interface mode before assuming a firmware issue, this is a genuinely newer product with fewer verified troubleshooting threads online than an established sensor.
  • Results seem inconsistent, confirm the module's pre-trained model actually matches what you are pointing it at, a person-detection model will not reliably classify objects it was never trained to recognize.

Optional Upgrade

Add the OLED display from the parts bin to show the vision module's current result live, turning this into a small standalone readout instead of something that only prints to a serial console.

Wiring

  • Grove Vision AI Module V2 VCC (Grove connector pin 1) → ESP32-C3 SuperMini 3V3 (or 5V if the module's Grove connector is 5V-tolerant per its datasheet — confirm before powering)
  • Grove Vision AI Module V2 GND (Grove connector pin 2) → ESP32-C3 SuperMini GND
  • Grove Vision AI Module V2 signal pin 1 → ESP32-C3 SuperMini GPIO8 — SDA if operating in I2C mode
  • Grove Vision AI Module V2 signal pin 2 → ESP32-C3 SuperMini GPIO9 — SCL if operating in I2C mode

The Grove connector itself (4-pin: VCC/GND + 2 signal lines) is a real, well-documented Seeed standard, so the power/interface wiring above is solid. The Grove Vision AI Module V2's exact command protocol and firmware API is a different matter — Seeed's Grove system supports both I2C and UART signaling over the same 4-pin connector depending on the module's mode/firmware, and which mode your specific module/firmware defaults to isn't something to assume. Rather than guess at a protocol, the wiring above defaults to I2C (using the ESP32-C3's default I2C pins) — confirm the correct mode for your module against Seeed's own Grove Vision AI V2 wiki before relying on it. If your module is set to UART mode instead, use two free UART-capable GPIOs for TX/RX instead of GPIO8/9.

For the actual model-inference commands and any request/response protocol with the module, defer to Seeed's official SenseCraft AI platform and firmware examples for the Grove Vision AI Module V2 — those are the authoritative, correct source for this hardware, not invented protocol bytes.

Wiring Diagram

VCC → 3V3 GND pin 1 → GPIO8 pin 2 → GPIO9 ESP32-C3 SuperMini Grove Vision AI Module V2

Starter Code

// Minimal link establishment only. This does NOT attempt to speak the Grove
// Vision AI Module V2's inference protocol — those exact command bytes are
// firmware/mode specific and should come from Seeed's official SenseCraft AI
// examples/wiki, not from an invented implementation here.

#include <Wire.h>

#define VISION_SDA 8
#define VISION_SCL 9
#define VISION_I2C_ADDR 0x62 // placeholder — confirm your module's actual address in Seeed's docs before use

void setup() {
  Serial.begin(115200);
  Wire.begin(VISION_SDA, VISION_SCL);

  Wire.beginTransmission(VISION_I2C_ADDR);
  uint8_t result = Wire.endTransmission();

  if (result == 0) {
    Serial.println("Grove Vision AI Module V2 responded on I2C — wiring/link OK");
  } else {
    Serial.println("No response — check wiring, power, and whether the module is set to I2C or UART mode");
  }

  Serial.println("Next step: use Seeed's official SenseCraft AI / firmware examples for model inference commands.");
}

void loop() {
  delay(5000);
}

Not sure about a step? Ask us

Try this next

Small plastic project enclosure with AHT20 sensor and ESP32-C3 breadboard, on a windowsill Build a Room Temperature & Humidity Monitor WS2812B LED strip glowing warm on a desk edge, wired to PIR sensor and ESP32-C3 on a breadboard Build a Motion-Activated Desk Light Small enclosure and breadboard with ESP32-C3, AHT20, SD card module and DS3231 RTC, laptop showing logged data graph Build a Wireless Environmental Data Logger