Dynamic adjustment of autonomous vehicle system based on deep learning optimizations

This invention describes a system for an autonomous vehicle (AV) to manage its internal deep learning systems. It works by first assessing how complex the driving environment is. If the environment is simple enough, the AV can reduce the size of its neural network, then test this smaller network in a simulation. Based on the simulation results, the AV adjusts its settings, and if performance is still not good enough, it can make the neural network even smaller.