Startup unveils first full brain emulation controlling a simulated body

Yesterday 09:50

By: Dakir Madiha

Startup unveils first full brain emulation controlling a simulated body

Eon Systems, a startup based in San Francisco, has introduced what it describes as the first full brain emulation of a fruit fly controlling a physically simulated body, producing several natural behaviors without machine learning training. The announcement, made on March 6, marks a significant step in brain emulation research by linking a complete neural map with an active body capable of movement.

The system builds on research published in Nature in 2024 by Eon scientist Philip Shiu and collaborators. Their earlier work created a computational model of the adult fruit fly brain, known as Drosophila melanogaster, containing more than 125,000 neurons and about 50 million synaptic connections. The model relied on the FlyWire connectome dataset and machine learning predictions of neurotransmitter identity.

That earlier version could predict motor neuron activity with about 95 percent accuracy but lacked a physical body to interact with. The new demonstration connects the connectome based brain model to a physically simulated fly body using the MuJoCo physics engine and the NeuroMechFly v2 biomechanical framework.

In the system, sensory signals flow into the digital brain. Neural activity then propagates through the entire connectome and produces motor commands that drive movement in the simulated body. The virtual fly performs actions including walking, grooming and feeding. The feedback loop between perception and motion is generated by the dynamics of the brain circuits themselves rather than reinforcement learning or scripted animation.

Researchers say this approach differs from earlier projects that simulated either neural systems without bodies or bodies controlled by artificial intelligence rather than biological brain structures. Previous work by DeepMind and the Janelia Research Campus produced a MuJoCo simulated fly controlled through reinforcement learning. The OpenWorm project attempted a similar embodied simulation of the worm C. elegans, which has a nervous system of 302 neurons.

Eon’s model instead reconstructs the wiring of a biological brain neuron by neuron using electron microscopy data. A separate effort by researchers at Sandia National Laboratories implemented the same FlyWire connectome on Intel’s Loihi 2 neuromorphic hardware. That system achieved major speed improvements compared with conventional simulation while reproducing similar neural behavior.

Eon says its long term goal is to scale the approach from the fruit fly brain to the mouse brain, which contains roughly 70 million neurons, and eventually to a human scale brain emulation.

To achieve this, the company combines expansion microscopy techniques for mapping neural connections with large scale datasets from calcium and voltage imaging to capture how neurons activate in living tissue.

Analysts following the field have noted that the fruit fly simulation has produced unexpectedly strong results for a relatively simple neural model, although important biological questions remain unresolved. These include how learning rules operate in real neural systems and the role of hormones, peptides and glial cells in neural computation.

Alex Wissner Gross, co founder of Eon Systems, wrote that if a fruit fly brain can now close the sensorimotor loop in simulation, scaling the technology to a mouse brain becomes primarily a question of computational scale rather than a fundamentally new problem.