Machine LearningTFLiteESP32Edge AI

Running ML Models on Microcontrollers with TFLite

February 10, 202510 min read

Why Edge AI?

Sending all sensor data to the cloud for processing introduces latency and requires constant connectivity. Edge AI solves this by running models directly on the microcontroller.

The Challenge

ESP32 has limited resources:

520 KB SRAM

4 MB Flash (typical)

240 MHz dual-core processor

Fitting a meaningful ML model in these constraints requires careful optimization.

Model Development

I trained an anomaly detection model in Python using TensorFlow:

python

import tensorflow as tf

model = tf.keras.Sequential([
    tf.keras.layers.Dense(16, activation='relu', input_shape=(5,)),
    tf.keras.layers.Dense(8, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid')
])

model.compile(optimizer='adam', loss='binary_crossentropy')
model.fit(X_train, y_train, epochs=50, batch_size=16)

Conversion to TFLite

python

converter = tf.lite.TFLiteConverter.from_keras_model(model)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
tflite_model = converter.convert()

with open('model.tflite', 'wb') as f:
    f.write(tflite_model)

The final model size was only 2.3 KB — perfect for ESP32.

Results

Inference time: < 5ms per prediction

Accuracy: 94.2% on test data

Power consumption: negligible increase

Conclusion

Edge AI on microcontrollers is not just possible — it's practical. The key is keeping models small and focused on specific tasks.

Back to all articles