notesum.ai

Published at December 9

Vision-Based Deep Reinforcement Learning of UAV Autonomous Navigation Using Privileged Information

cs.RO

cs.CV

cs.LG

Released Date: December 9, 2024

Authors: Junqiao Wang¹, Zhongliang Yu², Dong Zhou¹, Jiaqi Shi¹, Runran Deng³

Aff.: ¹Harbin Institute of Technology; ²Chongqing University; ³Individual

Arxiv: http://arxiv.org/pdf/2412.06313v1

Refer to caption

Operator	Input	Filters	Output
Conv2D	$1\times 80\times 100$	$3\times 3,8$ , stride 1	$8\times 80\times 100$
Max Pooling	$8\times 80\times 100$	$2\times 2$ , stride 2	$8\times 40\times 50$
Conv2D	$8\times 40\times 50$	$3\times 3,16$ , stride 1	$16\times 40\times 50$
Max Pooling	$16\times 40\times 50$	$2\times 2$ , stride 2	$16\times 20\times 25$
Conv2D	$16\times 20\times 25$	$3\times 3,25$ , stride 1	$25\times 20\times 25$
Max Pooling	$25\times 20\times 25$	$2\times 2$ , stride 2	$25\times 10\times 12$
Global Avg Pooling	$25\times 10\times 12$	$10\times 12$	$25\times 1\times 1$
Squeeze	$25\times 1\times 1$	-	$25$
State Feature	-	-	$8$
Concatenate	-	-	$33$
Fully Connected	33	128	128
Fully Connected	128	128	128
Fully Connected	128	4	4