Research Overview

The Evolving AI Lab focuses on evolving artificial intelligence, typically in robots and agents in simulated worlds. We study how evolution produced the complex, intelligent, diverse life on this planet by trying to computationally recreate it. A major focus is on evolving large-scale, structurally organized neural networks (i.e. networks with millions of connections that are modular, regular, and hierarchical). We are additionally interested in combining neuroevolution with learning algorithms (Hebbian, neuromodulation, etc.). We also investigate other bio-inspired AI techniques, such as deep learning.

For more information, please read our publications, press articles about the work, or watch a video of a talk where lab director Jeff Clune summarizes some of our recent research in this area. Other videos about our work are available here.

These are some keywords that describe related fields: evolutionary algorithms (also known as genetic algorithms or evolutionary computation), neural networks (including evolving neural networks, having them learn, deep learning, and computational neuroscience), robotics, artificial intelligence, and research into the evolution of intelligence, complexity, evolvability, and diversity.

We harness evolution and machine learning to have robots learn to walk, recover from damage, and adapt to unforeseen circumstances.

We investigate deep learning, which is a way to train deep neural networks (neural networks with many layers) to solve complicated tasks.

Our work shows that networks evolve to be modular when there is a cost for network connections: such modularity improves evolvability.

Creating computational evolutionary processes that--like natural evolution--endlessly produce a diversity of complex, interesting things.

Combining evolution with concepts from developmental biology produces repeated patterns in neural networks that increase intelligence.

Providing evolution with a larger palette of materials to work with leads to amazingly fun, quirky, lifelike virtual creatures.

Our work shows that we can make smarter robots by encouraging them to think differently about their environment and to try new things.

We investigate open questions in evolutionary biology with computational simulations of evolution.