Adaptive data transfer methods via policy evolution for UAV swarms
Phillip Smith, Robert Hunjet, Aldeida Aleti, Jan Carlo Barca
Abstract
Phillip Smith, Robert Hunjet, Aldeida Aleti, Jan Carlo Barca
Abstract
This paper presents an adaptive robotic swarm of Unmanned Aerial Vehicles (UAVs) enabling communications between separated non-swarm devices. The swarm nodes utilise machine learning and hyper-heuristic policy evolution to provide agility within the swarm, enabling each swarm member to select the most appropriate mobility policy for the environment given the swarm's abilities. The swarm evolution process of this study is found to successfully create different data transfer methods depending on the separation of non-swarm devices and the communication range of the swarm members. These methods are either human-designed, which the swarm adopts when most appropriate, or are novel hybridisations that the swarm creates for the problem. This paper also tests the swarm with individuals being removed during deployment. It is found that the swarm is immune to most alterations, though removal of specialised members of the heterogeneous swarm leads to temporary failure. The swarm evolution can then correct this failure by adjusting the swarm behaviour.
OpenAlex reports 9 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
This paper presents an adaptive robotic swarm of Unmanned Aerial Vehicles (UAVs) enabling communications between separated non-swarm devices. The swarm nodes utilise machine learning and hyper-heuristic policy evolution to provide agility within the swarm, enabling each swarm member to select the most appropriate mobility policy for the environment given the swarm's abilities. The swarm evolution process of this study is found to successfully create different data transfer methods depending on the separation of non-swarm devices and the communication range of the swarm members. These methods are either human-designed, which the swarm adopts when most appropriate, or are novel hybridisations that the swarm creates for the problem. This paper also tests the swarm with individuals being removed during deployment. It is found that the swarm is immune to most alterations, though removal of specialised members of the heterogeneous swarm leads to temporary failure. The swarm evolution can then correct this failure by adjusting the swarm behaviour.
Key concepts: Swarm behaviour, Swarm robotics, Computer science, Heuristic, Swarm intelligence, Process (computing), Coronavirus disease 2019 (COVID-19), Software deployment