Emergent Collective Intelligence in Swarm Robotics via Stigmergic Multi-Agent Reinforcement Learning

Overview

Individual ants possess limited cognitive capability, yet collectively they accomplish complex tasks without central command. This is enabled through stigmergy, an indirect, decentralized coordination via shared environment modification such as pheromone trails. Inspired by my own observations of carpenter ants, this project investigates a novel algorithm for achieving decentralized artificial collective intelligence via stigmergy. A custom simulation was built to train agents through a 17-stage curriculum of increasingly challenging tasks using multi-agent reinforcement learning, coordinating through a shared digital pheromone field. Trained policies were then transferred onto a custom-built physical robot swarm, demonstrating coordinated behaviour in the real world. Experiments confirm that stigmergy is causally beneficial, with pheromone-trained swarms significantly outperforming no-pheromone swarms across all swarm sizes. This open-source platform contributes to Physical AI, with applications in disaster response, planetary exploration, asteroid mining, and ecological monitoring, where resilient, decentralized systems are essential.

Awards (2)

• Bronze Medal
• Selected for CWSF 2026

Competition history

• CWSF 2026 Digital Technology

Related projects

Closest projects by meaning, across every fair and year in the corpus.