Skip to main content
. 2021 Jan 20;7:586707. doi: 10.3389/frobt.2020.586707

Figure 7.

Figure 7

The diagram shows the intended architecture for the DHG. The input is a tensor containing the data from different sources (section 2.2). The LSTM is unrolled over e previously generated data. The cell state h is the input of its corresponding unrolled unit in the next layer. However, using the latest LSTM unit's output, the DHG squeezes the prediction vector through a dense layer. Finally, the output of the network is a vector with three elements corresponding to forces in x, y, and z direction.