Convolutional Neural Networks (CNNs)
- Use: Lane, vehicle, object, and traffic-sign recognition.
- Example: CNN-based TSR and object-detection modules in production ADAS; traffic-sign recognition is implemented with CNNs in many OEM systems (Audi, BMW, Mercedes, Tesla, etc.). ResearchGate+1
RNN / LSTM
- Use: Time-series modelling for driver intent, vehicle-component health, sensor prediction.
- Example: RNN/LSTM-style models are described in EV battery predictive-maintenance patents used in electric vehicles to analyse sensor histories and predict failures. Google Patents
MLP / Perceptrons
- Use: Simpler signal classification (e.g., traditional ECUs, early driver-monitoring modules).\
- Example: Classic Driver Monitoring System (Lexus GS 450h and later models) uses neural-network style classifiers on camera features to determine driver attention and trigger warnings / braking. Wikipedia
Backpropagation Networks
- Role: Training algorithm used for practically all above neural-network models (CNN, RNN, MLP etc.).
- Note: Backprop is a generic training method, not a separate production feature, but underlies nearly all perception networks in modern ADAS.