Evolutionary strategies in swarm robotics controllers

• Autor(a) principal: Rodrigues, Alexandre Valério
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/27539
• topic: Evolutionary algorithms; Unmanned vehicles; Realistic simulations; Behavior tree controllers; Algoritmos evolucionários; Veículos não tripulados; Simulações realistas; Controladores com behavior trees

Nowadays, Unmanned Vehicles (UV) are widespread around the world. Most of these vehicles require a great level of human control, and mission success is reliant on this dependency. Therefore, it is important to use machine learning techniques that will train the robotic controllers to automate the control, making the process more efficient. Evolutionary strategies may be the key to having robust and adaptive learning in robotic systems. Many studies involving UV systems and evolutionary strategies have been conducted in the last years, however, there are still research gaps that need to be addressed, such as the reality gap. The reality gap occurs when controllers trained in simulated environments fail to be transferred to real robots. This work proposes an approach for solving robotic tasks using realistic simulation and using evolutionary strategies to train controllers. The chosen setup is easily scalable for multirobot systems or swarm robots. In this thesis, the simulation architecture and setup are presented, including the drone simulation model and software. The drone model chosen for the simulations is available in the real world and widely used, such as the software and flight control unit. This relevant factor makes the transition to reality smoother and easier. Controllers using behavior trees were evolved using a developed evolutionary algorithm, and several experiments were conducted. Results demonstrated that it is possible to evolve a robotic controller in realistic simulation environments, using a simulated drone model that exists in the real world, and also the same flight control unit and operating system that is generally used in real world experiments.