Roman Miletitch's research while affiliated with Université Libre de Bruxelles and other places

... However, although the response threshold model improves foraging efficiency, it does not consider physical interference when seeking to improve TA performance. As has been noted elsewhere, the performance of swarm-robotics foraging is influenced by the physical interference among the active foraging robots [30,31]: having too many robots foraging simultaneously leads to less efficient foraging [32,33]. However, previous studies on TA did not take physical interference into account when adjusting the number of active foraging robots. ...

... Developing swarm robotics collective decision-making algorithms and mechanisms may be an important area of research in the future. This involves developing information sharing strategies and coordination algorithms, designing cognitive decision frameworks, and allocating resources and tasks so that swarm robots can effectively collaborate and complete complex tasks [21,158]. This may include self-organization, self-configuration, and self-healing techniques within the swarm. ...

... Foraging can be studied in various forms depending, for example, on how resources are scattered in the environment or the number of target locations 28 , and foraging studies can focus on various different aspects, for example, on the coordinated navigation of the environment 29,30 , on the collective transport of resources 31,32 , or on the allocation of robots between exploratory and exploitative tasks 33 . Here, we investigate central place foraging, where robots have to transport resources (also named food) to a single central depot location (named nest) 34 . ...

... In the modular approach, preexisting software modules are combined and tuned by an optimization algorithm. Results show that modular methods are more suitable to produce communication-based behaviors [14] and are more robust to the so-called reality gap [8,15], that is, the possibly subtle but unavoidable differences between reality and the simulation models used in the design process. ...

... Every 30 seconds, the robot reads the last message received from its neighbours, that used to update its opinion through one of the two state machines of Fig. 1. As shown in previous research 34,54 , using a relatively low frequency of robot's opinion update (30 s) compared with the random diffusion speed (1 cm/s) allows obtaining a qualitative good agreement between the macroscopic modelswhich assume a well-mixed interaction topology-and the robotic implementation-which relies on local interactions in a range of 10 cm. The robots showed their opinion via the coloured LED (option X as red, option Y as blue, and no opinion as green). ...

... Aplicado principalmente en estudios de sistemas multi-agente [15] [16], en este método los agentes aprenden a través de interacciones por prueba y error con su entorno, recibiendo una retro-alimentación positiva o negativa según su desempeño. Recientemente se ha propuesto Demo-cho [5], un método de diseño automático que combina el aprendizaje por refuerzo inverso [17] con diseño automático modular (AutoMoDe-Chocolate [18]) para generar comportamientos de una forma mas intuitiva. Demo-Cho sólo requiere de demostraciones del comportamiento deseado para aprender y generar el software control correspondiente. ...

... While simple living organisms relied on stigmergic communication for coordination, artificial swarms can also relatively easily exchange direct messages with structured content, potentially simplifying their deployment in the real world. Through direct local communication, robot swarms can coordinate and exploit a form of social odometry in order to efficiently navigate through the environment without GPS 29,[43][44][45] . The robots, through what is often described as the "many-wrongs principle" 46,47 , compensate for individual odometry errors which are filtered out by mechanisms of information pooling and achieve higher navigation accuracy as a group than they would do as single individuals. ...

... Besides neuro-evolution, other approaches have been proposed. For example, Francesca et al. (2014a) used probabilistic finite-state machines as a control architecture for the individual robots, while Jones et al. (2016) and Kuckling et al. (2018) used behavior trees. ...