A Survey on Automatic Design Methods for Swarm Robotics Systems

Abstract

Swarm robots are a branch of robotics that draws inspiration from biological swarms to mimic their collective behavior. Automatic design methods are part of swarm engineering, depend on artificial intelligence algorithms to produce the collective behavior of robots. In general, they follow two-approach evolutionary algorithms like practical swarm optimization and reinforcement learning. This paper studies these approaches, illustrating the effect of modifications and enhancements of algorithms for both directions, showing important parameters considered for the best performance of the swarm, and explaining the methods and advantages of using deep learning to reinforcement learning.

Curriculum learning for deep reinforcement learning in swarm robotic navigation task

“…The behavior-based design method explicitly defines rules for each robot, dictating their actions to achieve the desired collective behavior. On the other hand, the automatic design method utilizes algorithms rooted in artificial intelligence, enabling the robots to become intelligent entities capable of learning and adapting from their environment (Iskandar et al, 2021) Among the automatic design methods, two primary approaches have gained significant attention in SR: evolutionary algorithms and reinforcement learning (RL). Evolutionary algorithms, such as Particle Swarm Optimization (PSO) (Gbenga et al, 2016) and Bacterial Foraging Optimization Algorithm (BFOA) (Yang et al, 2014), rely on computational models inspired by natural evolution to optimize the collective behavior of the swarm.…”