Evolution of behaviour trees for collective transport with robot swarms

Swarm robotics, inspired by natural swarms, studies how simple robots with only local sensing capabilities and no centralised control may cooperate to achieve a common goal in a robust, flexible and scalable way. A robotic system with such properties constitutes an interesting alternative to the platforms currently used in warehouses and distribution plants, where workers are at risk of injury and the space and budget available for complex infrastructure is limited. Swarm behaviours are emergent, which makes the task of designing the controllers of the individual robots particularly challenging. In this work, we propose a method for a swarm of industrial robots to collectively transport items that are too heavy for a single agent to carry. We use artificial evolution to evolve behaviour tree controllers for the swarm agents and we conceive a decentralised coordination strategy based on local messaging. The method is developed and tested in a simulated environment, using a combination of freely available open source libraries. The results show that a homogeneous swarm equipped with our solution is able to successfully find the items placed in the environment and transport them back to a nest region. We suggest further tuning of the evolutionary parameters and the introduction of noise in the simulator in order to improve the observed performance of the controllers in simulation and their expected performance the real worldObjectius de Desenvolupament Sostenible::9 - Indústria, Innovació i Infraestructur