Animal Intelligence Group

Animals solve physics problems that stump the best AI systems. We're figuring out how.

Janelia Research Campus, HHMI

🏗️

Robotic Behavioral Arenas

We build robotic systems that create complex dynamic environments for rats, mice, fish, and flies — then observe how animals move, make decisions, learn, and reason about the physical world.

🦴

Biomechanical Digital Twins

We build increasingly realistic musculoskeletal models of the rat, mouse, fish, and fly in MuJoCo. These digital twins of animal bodies are the foundation for training embodied AI that moves and acts like real animals.

🤖

Embodied AI

We train virtual animals to imitate real animal behavior and cognition in physics simulations. By solving the same problems under the same physical constraints, our models converge on similar computational strategies as real brains.

🧠

Connectomes & Neural Models

For fish and flies, we collaborate with partner labs to simulate their connectomes controlling virtual bodies. For rodents, we develop more abstract AI models that learn to sense, act, and plan like real animals.

Intelligence is a general attribute that allows animals to build world models — internal representations of how the world operates and how they can interact with it to achieve distant goals. Animals evolved to solve a variety of problems under strong pressure to be efficient, and this convergent evolution in problem-solving space is a powerful guide for building AI.

Our approach: by training digital twins of animal bodies in increasingly realistic physics simulations, we expect our models to learn control strategies and representations that parallel those found in real nervous systems. This makes embodied AI a tool for neuroscientific discovery — reverse-engineering intelligence by rebuilding it.

We work at Janelia Research Campus, HHMI, combining robotics, biomechanical modeling, machine learning, and behavioral observation.

Contact

Rob Johnson Group Leader, Mechanistic Cognitive Neuroscience Janelia Research Campus, HHMI 19700 Helix Drive, Ashburn, VA 20147

Bluesky · Google Scholar · GitHub · YouTube