Unmanned vehicle local autonomous control method, device and equipment based on depth map

The invention relates to an unmanned vehicle local autonomous control method, a device and equipment based on a depth map. The method comprises the following steps: acquiring the depth map shot in the field of view of an unmanned vehicle and extracting a depth feature vector of the depth map; splicing and fusing the depth feature vectors corresponding to a plurality of depth images continuously shot at historical moments and the navigation target point position coordinates of the unmanned vehicle when the depth images are shot to obtain a fused feature vector, and taking the fused feature vector as the input state of the navigation neural network of the unmanned vehicle; designing a comprehensive reward function; training a navigation neural network by adopting a hyper-parameter segmentation training strategy in an obstacle simulation environment by utilizing the fusion feature vector and the comprehensive reward function; and in a real physical environment, processing a depth image by using the trained navig