Artificial Dreamers by Jake O’Brien

Artificial Fringe


Artificial Fringe is an experiment in using machine learning and AI to do something very different than those tools are typically used for: creating autonomous agents that express themselves and communicate with those around them.

Machine Learning is generally used to create agents that are good at solving specific problems, like spotting spam emails or enabling robots to stand up and walk. AI in games is usually used to create computer opponents that are good at winning and therefore challenging to humans.

This project uses those tools to create a simulated dance.

Each dancer’s dance is influenced by a variety of things. When they ignore their neighbors completely, they are dancing with a neural network trained on 3d animations of dance. None of what they are doing is hand-created. It’s all procedurally generated by the neural network. No dance move is the same.

But they also watch their neighbors and imitate. Sometimes they imitate multiple neighbors. So generally their movement is a blend what their own brains (hearts?) are telling them and what is happening around them. It’s like behavioral flocking. You can see little pockets of a specific dance appear and disappear on the dance floor.

In addition to this, they have other behaviors that they will either start randomly or based on circumstances. They may engage with another dancer, they may follow others, they look at each other, and they move around each other with actual boid flocking.

Also, you can let balls loose and they will chase and hit them around.

They are all subject to physics. They are not ragdolls, but they have full collider rigs and IK solvers and are moved via physics.

All of these layered systems combine to make pretty believable dance floor sim.