Figure 2.

Proposed convolutional neural network based navigation framework. It consists of four convolutional layers, three pooling layers, and two fully connected layers. The inputs are raw, uncalibrated spherical images and the outputs are navigation signals (steering direction and angles).