Reinforcement Learning with TensorFlow

Sensor fusion

This step includes fusing the inputs from different sensors and processes, and preparing them to feed into the deep neural network. Each sensor information captures the state of the environment in the form of a raw vector. Grouping all these raw vectors is done and fed into a deep neural network. Each sensory input will form a separate feature vector. Thus, as a result of learning, that is, cost minimization, optimization of the weights associated with each of those sensor features occurs. These learned weights quantify the relevancy of the corresponding sensor features. As far as the deep neural network is concerned, CNN is the best choice for the task.