Firstly, a fully convolutional residual network [28] was implemented to predict depth information from the input monocular RGB images. The estimated depth images were subsequently fed into a dueling architecture based double DQN for action selection. Due to the extraordinary performance of Convolutional...
Firstly, a fully convolutional residual network [28] was implemented to predict depth information from the input monocular RGB images. The estimated depth images were subsequently fed into a dueling architecture based double DQN for action selection. Due to the extraordinary performance of Convolutional...