Janelia and Google DeepMind researchers have created a virtual fruit fly. Janelia has shared the news in a press release, and DeepMind in a thread posted to X.

This digital insect walks and flies like a real one. The researchers used artificial intelligence (AI) to make it lifelike.

The virtual fly has an accurate body model. It copies a fruit fly’s outer skeleton and movements. The researchers trained an artificial neural network to control it. They used real fly behavior videos to teach it. The virtual fly can follow complex flight paths. It even uses its eyes to steer while flying.

The researchers took real fly data on walking and flying. They trained the network to mimic these actions. Then, the network controls the virtual fly’s movements. The researchers compared it to a tiny brain for the fly. The model uses MuJoCo, a physics simulator for robotics. A physics simulator creates realistic movements in a virtual world. The researchers added features like fluid forces on wings. This helps the fly model fly properly. They also added adhesion actuators. Adhesion actuators copy how insect feet grip surfaces.

Advancing Neuroscience Research

This virtual fly helps scientists understand behavior better. It shows how the brain, body, and environment work together. Labs often struggle to measure these interactions. The model reveals details like forces during flight. The researchers said the simulation explains how brain commands turn into actions. The fly’s body shape and physics play a big role. The researchers built on past models like NeuroMechFly. NeuroMechFly had a realistic body for walking simulations. The new model improves anatomy and behavior details. It can walk at 2 cm/s while turning. It also flies at 30 cm/s with evasive maneuvers.

The researchers plan to make the fly more realistic. They want to add muscles, tendons, and a sensory system. They also aim to use a real fly neural network. Future projects include virtual mouse and zebrafish models. They shared their methods openly for other scientists.

The researchers have described the methods and results of this study in a preprint published in bioRxiv. This preliminary paper also appears in Nature, with the disclaimer "We are providing an unedited version of this manuscript to give early access to its findings."