• Citer

Automatic design of pheromone-based stigmergy in robot swarms

par Salman, Muhammad
Président du jury Dorigo, Marco
Promoteur Birattari, Mauro
Publication Non publié, 2024-08-30

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

This dissertation aims to automatically design pheromone-based stigmergy in robot swarms, addressing two key challenges in swarm robotics: i) the automatic design of stigmergy-based collective behaviors, and ii) the real-world implementation of such systems. While manual design of stigmergy-based behaviors in swarm robotics is possible, it is a complex and resource-intensive process. Designing collective behaviors for loosely coupled, autonomous robots is inherently difficult, and even more so when environmental modification is necessary. Optimization-based design provides an alternative, but could often requires some human input to fine-tune a produced control software. This dissertation explores a fully-automatic design method that is based on automatic modular design method, striving to eliminate the need for human intervention. While stigmergy-based behaviors hold great promise for robot swarms, practical implementation faces challenges. Virtual stigmergy aids in simulation, but fails to capture the dynamic, environment-altering nature essential to its true potential. Physical methods like alcohol trails or external tracking systems can be expensive, complex, and potentially compromise robot autonomy. This research introduces Phormica, a cost-effective solution using special surfaces and UV LEDs to create artificial pheromones, enabling broader experimentation in real-world scenarios. To demonstrate these concepts, I introduce Habanero, an automatic design method to design stigmergy-based control software. Using a swarm of e-puck robots, I evaluate Habanero’s performance across different missions and compared it to alternative design methods like neuroevolution-based design, manual design, and random-walk behavior. Our experimental results show that Habanero is a viable approach for pheromone-based stigmergy design. The produced behaviors are comparable to, sometimes outperforming, those created by human designers. Interestingly, while Habanero’s modules are mission-agnostic, the devised strategies are mission-specific. The automatic process effectively leverages pheromone-based stigmergy to create collective behaviors demonstrating spatial organization, memory, and communication.