Work

Morphogenesis

Multi-agent swarm intelligence with emergent role specialization. 64 decentralized agents learn to cooperate via attention-based communication — developing scout, gatherer, and defender roles without any hand-coded behaviors.

A side-by-side comparison of classical boids (3 hand-coded rules) versus a trained MAPPO policy (250K parameters) on identical environments with adversarial threats. Watch the RL agents outperform in real-time.