Episode Thirty-Four: Sabine Hauert

Engineering & simulationartificial evolution

The team uses artificial evolution to automatically design individual agents and their local interaction rules, optimizing them to achieve desired collective swarm behaviors without manual coding.

“But increasingly, we use machine learning, right? To do this automatically. So we use a tool called artificial evolution, and that allows us to automatically design individuals, their local interactions, and we give them points when they achieve the desired collective swarm behavior, and so we can do this automatically increasingly.”

Engineering & simulation

The team uses machine learning to prioritize operator-defined characteristics (like safety or energy usage) to automatically generate and deploy controllers for swarms that balance these trade-offs.

“And what you do as a swarm operator, and one of the latest papers that we have is you prioritize the characteristics you care about. And what comes out of it is an automatically generated controller for the swarm that gets deployed that has those trade-offs.”