Authors: Zhe Ren, Xinghua Li, Yinbin Miao, Zhuowen Li, Zihao Wang, Ximeng Liu, Robert H. Deng Keywords: UAVs, Gossip Protocol, Sparse Rewards, Partially Observable Markov Decision Process, Reinforcement Learning

• Python
• NS-3 (version == 3.25)

The Proof-of-concept is made up of two parts as follows:

• UAVNET environment simulated using NS-3
• Deep Reinforcement Learning code

Among them, the UAVNET environment code is GossipEnvironment.py, while others compose the deep reinforcement learning code.

Note: The NS-3 simulation environment with python support needs to be installed, more details see: ns-3 | a discrete-event network simulator for internet systems (nsnam.org)

Python Dependencies:

• pytorch
• gym
• tqdm

• After alter the train parameters, to train a new model:

$ python branching_dqn.py

• Using a pre-trained model:

$ python enjoy.py