Neuro-Evolutionary Swarms

Background: Initially, the project was supposed to be a simple understanding and implementation of existing static swarm behaviors studied by others. Later I grew curious to see if there was a way to evolve dynamic behaviors just like nature does. I understood the power of neural networks and paired it with genetic algorithms to see if a dynamic robot controller could result in complex behaviors. I tested out a hypothesis that inter-dependent evolutionary pressures drives the complex behavior of communication to evolve in animals. I successfully managed to develop a nascent form of communication, where prey robots where able to alert other prey robots about predator robots in the vicinity.

Experiments: ARGoS a multi-robot simulator was used to conduct experiments on Foot-bot robots (part of Swarmanoid project) which had following capabilities:

1. RGB LEDs
2. Omnidirectional camera
3. Proximity sensor

It was observed that prey bots rotated about their axis and maintained equal distance from each other so as to increase the scope of detecting a predator bot. The predator bot evolved to chase a prey that was closest to it. The preys evolved a nascent communication whereby they started blinking whenever they were chased by a predator and were in the vicinity of other preys. Any prey bots which detected the signal moved away quickly and did the same to notify other prey bots. The experiments gave a proof of the earlier hypothesis and showed that neuro-evolutionary models can be used to build complex behaviors