Abstract: This research project investigates how a new bee-inspired Swarm intelligence algorithm, i.e. Stigmergic Landmark Foraging (SLF), can be directly deployed on a swarm of robots (e-puck). The algorithm has been shown to be robust, scalable and efficient in simulation. Furthermore, this proposal describes a set of experiments to investigate how well the algorithm is capable to coordinate a large collective of robots (up to 120) in a situated foraging task. Robustness, efficiency and scalability will be tested.
Pursuing this research agenda implies an important breakthrough in swarm robotics for two reasons, one, it would be the first time principles from swarm intelligence optimization are directly transferred to and explicitly used in multi-robot systems, and two, successful control and coordination of a large robot collectives is required to tackle contemporary technological challenges. Moreover, it would be the first successful large-scale coordination swarm robotics experiments in the Netherlands. The former point is crucial as approaches from the current state of the art are implicit, i.e., they use evolutionary algorithms to evolve a neural controller (encoding the swarm behaviour) for the robots. This approach demands quite some parameter fine-tuning of the algorithm, a good design of the fitness function and can be complex to scale. The latter point implies potential application to security patrolling, monitoring of environments, exploration of hazardous environments, search and rescue in crisis management situations and others.
The novelty of this project lies in the hybrid approach, which combines ideas from heromone-based algorithms with non pheromone based algorithms and Reinforcement Learning. The most investigated type of swarm algorithms are pheromone based, inspired by the biological behaviour of ants. Despite the success of this type of algorithms in software applications, they have not been directly transferred to the physical world of swarm robotics due to the technological challenge of deploying and sensing artificial pheromone. SLF is a hybrid approach, which combines the strong adaptability of ants using landmarks, with the efficiency of the bee path integration algorithm. More precisely, the algorithm constructs stigmergic landmark networks over which agents can travel from and to a goal. On top of this the algorithm incorporates an online reinforcement learning approach, combining the fields of Multi-agent Reeinforcement Learning and Swarm Robotics to a greater extent. As such we integrate individual learning, as encountered in RL, with the adaptability of the collective, reached through the mechanisms of Swarm Intelligence.