End-to-end Learning System Using Recursive Neural Networks

[Erica Zhu Feilong Jiangli]

Farreo has developed an end-to-end learning system for speed control. The system adopts the function of neuron network and long-term memory (LSTM), which is a recurrent neural network (RNN), and can learn long-term dependencies.

The new system was developed by Wirbel and her colleagues, who also deployed an artificial neural network (ANN), which relies on in-depth learning technology. The network has been provided with a large number of demonstrations of human drivers'vehicle operation, which are recorded by forward-looking cameras and can be closely related to the driver's visual field when driving.

The neuron network will also be trained to imitate the driver's behavior, especially focusing on reproducing the current speed of the vehicle. For example, when the input image contains a speed-limiting panel of 50 km/h, the neuron network will ensure that the current speed of the vehicle is not higher than 50 km/h.

Perhaps in the near future, the neuron network system will be deployed in the automatic driving vehicle to improve the efficiency of speed control and achieve more intuitive driving control. Researchers plan to apply this proof-of-concept to more complex driving situations, adding a variety of complex driving manipulations, such as diversion, turning at intersections or vehicle navigation at roundabouts.