AI and mechanics driving the rise of bio-inspired robotic swarms

Biological collectives such as slime molds, fungal networks, and bird flocks achieve robust coordination and adaptability through decentralized interactions among simple agents. This work explores how these principles can be transferred to engineered swarm robotic systems operating in complex environments. We present a data-driven framework that combines machine learning, including neural networks and reinforcement learning, with physics-based computational mechanics simulations to model agent-environment interactions under realistic conditions and generate rich training data. The learned decentralized policies are implemented in an in-house multiagent platform and transferred to physical swarm robot prototypes, enabling direct evaluation of sim-to-real performance. Results demonstrate improved robustness, scalability, and fault tolerance across tasks such as distributed sensing, adaptive exploration, and decentralized task allocation. Overall, this work shows that integrating biologically inspired organization with data-driven learning and mechanics-informed modeling enables resilient swarm robotic architectures suitable for real-world deployment.