Fig. 5: In situ training of an FE-PS-NET which acts as a “visual system” of a prototype autonomous vehicle.
From: In situ training of an in-sensor artificial neural network based on ferroelectric photosensors
a Photography of the vehicle, along with schematic circuit diagrams. The output current of the FE-PS-NET is fed to the neuron unit, and then sent to the motor system of the vehicle. b Four traffic signs used for training and test. c Flowchart of the in situ training of the FE-PS-NET. The operations in the light blue boxes are implemented in hardware, while those in the orange boxes are temporarily implemented in software on a PC and can be implemented by on-chip integrated circuits in the future. Evolutions of d MSE, e weights, and f normalized Iouts for different input traffic signs with the training epoch.
