Cultural evolution of perceptually grounded compositional lexicons in swarm robotics systems

Communication in swarm robotics has traditionally been addressed from the evolutionary robotics perspective. However, cultural evolution, language games, and cultural transmission offer promising alternatives for achieving richer and more complex emergent communication. Within this framework, the ability to self-organize and share a common lexicon to refer to relevant objects, events, or concepts in the environment is essential for language-like communication among robots. In this paper, we show that robot swarms can self-organize a compositional lexicon to refer to multiple types of objects, with all swarm members converging to the exact same set of words. Our proposed system is based on cultural evolution, language games, semiotics, and the principle of embodiment, ensuring that words and meanings are perceptually grounded in each robot’s sensory experience and interaction with the environment. We assess the scalability of the proposed system and demonstrate that swarms of up to 100 robots can achieve a common lexicon of 26 different words. We further demonstrate that compositionality of the lexicon is shared by all robots in the swarm, even though individual meanings and compositional rules remain private to each agent.