Neural Vehicle State Estimation Under Noisy Sensor Conditions

The patent replaces traditional mechanical/mathematical filtering systems (like Kalman filters) with an artificial neural network-based system. The ANN learns optimal state estimation directly from sensor data through training, substituting explicit mathematical models with a data-driven adaptive system that handles non-linearities and noise more effectively.

The patent changes the fundamental parameters of the estimation system by using learned parameters from training data rather than fixed mathematical models. The ANN adapts its internal parameters (weights and biases) based on training examples, allowing it to optimize state estimation accuracy for specific operating conditions and sensor configurations.

The patent performs the computationally intensive work of training the neural network offline before deployment. During actual vehicle operation, the pre-trained ANN processes sensor data much faster, as the heavy lifting of learning optimal estimation patterns has already been completed. This separates the training phase (high computational power) from the inference phase (lower computational power).

The patent uses approximate computation methods and simplified ANN architectures that provide sufficient accuracy for safety-critical applications without requiring excessive computational resources. The system accepts certain approximations in exchange for reduced processing demands during operation.