ESP32-S3 AI Robotics Platform
An autonomous, open-source micro-rover designed and built by the students of Elite Science Club. Featuring on-device INT8 convolutional neural network inference, deterministic FreeRTOS dual-core task dispatching, and closed-loop motor telemetry.

You Don't Need $1,000 to Build an AI Robot
Watch our 35-second demo on why edge AI robotics is accessible on a student budget.
1. System Architecture Breakdown
⚡ Dual-Core Compute & Neural Engine
The rover is powered by the ESP32-S3-WROOM-1 (N16R8), housing two 32-bit Xtensa LX7 cores running at 240 MHz. Unlike older microcontrollers, the ESP32-S3 includes dedicated vector instructions (PIE) that accelerate matrix multiply-accumulate operations, enabling quantized INT8 machine learning models to run locally without cloud latency or connectivity.
🔄 Real-Time FreeRTOS Task Separation
To prevent perception stalls from interrupting kinematic motor control, we pin workloads to specific cores:
- Core 0 (Perception Loop): Captures OV2640 camera frames, unpacks I2S audio buffers, and executes neural network forward passes.
- Core 1 (Kinematics & Telemetry): Runs the 100Hz closed-loop PID velocity controller, reads wheel encoders, and handles ultrasonic fail-safe polling.
👁️ Multi-Modal Edge Perception
Visual inputs flow through an OV2640 sensor directly into 8MB of Octal PSRAM via DMA. Acoustic keyword spotting is powered by an INMP441 omnidirectional MEMS microphonestreaming 24-bit audio directly over an I2S digital audio bus, eliminating analog noise interference.
🔋 Power Architecture & Filtering
Supplied by a 2S 7.4V Li-Ion battery pack (2000 mAh), the power rails are split into an unfiltered motor rail and an isolated 5.0V digital logic rail driven by a switching MP1584EN buck regulator. Decoupling capacitors (100nF ceramic + 470µF low-ESR electrolytic) isolate motor inductive kickback from resetting the processor.
2. Bill of Materials (BOM)
Complete open hardware specification. All components are standard, widely available commercial off-the-shelf (COTS) parts.
| Ref | Component | Technical Specification | Subsystem | Functional Role |
|---|---|---|---|---|
| U1 | ESP32-S3-WROOM-1 (N16R8) | Dual-core Xtensa LX7 @ 240MHz, 16MB Flash, 8MB Octal PSRAM, Vector Extensions | Compute & Processing | Central AI inference and control core |
| CAM1 | OV2640 Image Sensor Module | 2.0 Megapixel CMOS, 8-bit DVP parallel bus, UXGA / CIF support | Vision Perception | Captures frames for on-device object detection |
| MIC1 | INMP441 Omnidirectional MEMS Mic | Digital I2S output, 24-bit audio sampling, 61 dBA SNR | Acoustic Perception | Keyword spotting and acoustic command recognition |
| U2 | DRV8833 Dual H-Bridge Driver | Up to 10.8V, 1.5A per channel RMS, hardware current limiting & PWM | Motion & Actuation | Dual DC gear-motor differential drive control |
| M1-M2 | N20 Metal Geared DC Motors | 6V 300 RPM with integrated 12-CPR magnetic quadrature encoders | Motion & Actuation | Closed-loop wheel velocity feedback & movement |
| SEN1 | MPU6050 6-Axis Motion Sensor | 3-axis gyroscope, 3-axis accelerometer, 400kHz I2C interface | Navigation & Inertial | Heading tracking, tilt sensing, and rollover protection |
| SEN2 | HC-SR04P Ultrasonic Sensor | 3.3V compatible, 2cm to 400cm range, 3mm resolution | Obstacle Detection | Hardware fail-safe distance barrier detection |
| PWR1 | 2S 7.4V 18650 Li-Ion Pack (2000mAh) | 7.4V nominal, integrated BMS protection board (overcharge/discharge) | Power & Regulation | Primary energy reservoir for drive motors & electronics |
| VR1 | MP1584EN Step-Down DC-DC Converter | High-efficiency switching regulator, tuned to 5.0V @ up to 3A output | Power & Regulation | Steps down battery voltage to stabilized 5V logic rail |
3. Pinout & Hardware Bus Assignments
Carefully verified GPIO allocation ensuring zero pin strapping conflicts during ESP32-S3 boot cycle.
| Periphery Signal | ESP32-S3 Physical GPIO | Bus / Protocol |
|---|---|---|
| Camera Pixel Data [D0–D7] | GPIO 11, 12, 13, 14, 15, 16, 17, 18 | DVP Parallel |
| Camera XCLK & PCLK | GPIO 10 (XCLK), GPIO 9 (PCLK) | DVP Timing |
| Camera SCCB (I2C Config) | GPIO 21 (SDA), GPIO 22 (SCL) | I2C Bus 0 |
| INMP441 I2S WS (Word Select) | GPIO 4 | I2S Bus 0 |
| INMP441 I2S SCK (Bit Clock) | GPIO 5 | I2S Bus 0 |
| INMP441 I2S SD (Data Out) | GPIO 6 | I2S Bus 0 |
| DRV8833 Motor Left IN1 / IN2 | GPIO 1 & GPIO 2 (LEDC PWM) | LEDC Channel 0/1 |
| DRV8833 Motor Right IN3 / IN4 | GPIO 42 & GPIO 41 (LEDC PWM) | LEDC Channel 2/3 |
| MPU6050 IMU SDA / SCL | GPIO 8 (SDA), GPIO 7 (SCL) | I2C Bus 1 |
| Ultrasonic Trig / Echo | GPIO 38 (Trig), GPIO 39 (Echo) | General GPIO |
4. Firmware Architecture & Dual-Core Task Dispatch
Snippet from our open-source C++ firmware showing FreeRTOS core assignment and the quantized edge AI execution loop.
// Elite Science Club (ESC) - ESP32-S3 Edge AI Robotics Core
#include <Arduino.h>
#include "esp_camera.h"
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
// Task Handles for Dual-Core Execution
TaskHandle_t TaskPerceptionHandle = NULL;
TaskHandle_t TaskKinematicsHandle = NULL;
// Core 0: High-Compute Vision Perception & TinyML Inference
void PerceptionTask(void *pvParameters) {
for(;;) {
camera_fb_t *fb = esp_camera_fb_get();
if (!fb) {
vTaskDelay(pdMS_TO_TICKS(10));
continue;
}
// Run INT8 quantized CNN obstacle classifier using ESP32-S3 Vector PIE instructions
// Output: Vector of detected path centroids
esp_camera_fb_return(fb);
vTaskDelay(pdMS_TO_TICKS(33)); // ~30 FPS loop
}
}
// Core 1: Deterministic Real-Time Kinematics & Motor Control
void KinematicsTask(void *pvParameters) {
for(;;) {
// 100Hz PID wheel velocity closed-loop update
// Reads quadrature encoders and safety ultrasonic HC-SR04
vTaskDelay(pdMS_TO_TICKS(10)); // 100 Hz deterministic rate
}
}
void setup() {
Serial.begin(115200);
// Pin Perception Task to Core 0 (PRO_CPU)
xTaskCreatePinnedToCore(
PerceptionTask, "TaskPerception", 8192, NULL, 2, &TaskPerceptionHandle, 0
);
// Pin Kinematics Task to Core 1 (APP_CPU)
xTaskCreatePinnedToCore(
KinematicsTask, "TaskKinematics", 4096, NULL, 3, &TaskKinematicsHandle, 1
);
}
void loop() {
vTaskDelete(NULL); // Primary logic managed by FreeRTOS tasks
}Want to Reproduce This Build in Your School or Lab?
We share all schematics and firmware freely under open-source licenses as part of our"Science For Service" charter. Contact us for workshops or student guidance.