Research on multi UAV cooperative collision avoidance strategy based on deep reinforcement learning

DOI:	10.11809/bqzbgcxb2023.03.025
Keywords:	multiple UAVs; cooperative collision avoidance; deep reinforcement learning; IPPO; observation design
Abstract:	Aiming at the problem that multiple UAVs may collide with static obstacles or other UAVs during a cooperative mission, this paper proposes a cooperative collision avoidance strategy based on deep reinforcement learning. Firstly, each UAV is regarded as an independent decision making individual, and the deep neural network is applied to fit its strategy function and value function.Then, based on the Independent Proximal Policy Optimization(IPPO) Algorithm, an observation design method which only observes and ranks part of target information is proposed. It solves the problem that it is difficult to train the neural network when its input dimension is too large. Finally, by taking the collision avoidance problem of 25 UAVs in the process of a cooperative mission as an example, this paper carries out algorithm design and network structure design, and verifies the effectiveness of the proposed collision avoidance strategy through simulation experiments.
Published:	2023-03-28
PDF HTML