The Ant and the Trap: Evolution of Ant-Inspired Obstacle Avoidance in a Multi-Agent Robotic System

Abstract

In this paper, we demonstrate that a simple robotic swarm can use a genetic algorithm (GA) to achieve autonomous navigation in highly cluttered environments. We also show the GA can evolve effective strategies for coping with random obstacle placements using a central place foraging algorithm (CPFA). The CPFA combines stochastic movements, insect-inspired obstacle avoidance, and simple communication among robots. We modify the CPFA by allowing simulated robots to communicate with stygmergic trails in CPFA-trails (CPFAT). Both the CPFA and CPFAT can evolve foraging strategies to collect resources in the presence of a high density of randomly placed obstacles. CPFAT additionally succeeds against a classic bug trap, where the CPFA fails. These methods are simple to implement, run in real-time, and need no global knowledge of the environment.