Carbon-Based Computing Challenges Silicon-Based GPU? Full Drosophila Brain Computational Model Drives Robotic Body to Perform Natural Behaviors

According to 1M AI News's monitoring, the whole-brain simulation company Eon Systems recently demonstrated the first closed-loop system driving a physically simulated body with a complete brain simulation: a full-brain model based on a real fruit fly connectome drove a virtual fruit fly in the MuJoCo physics engine to perform walking, grooming, and other natural behaviors. No one had previously achieved a full brain simulation derived from a biological connectome to drive a physically simulated body to exhibit diverse natural behaviors.

The core of the system is Eon's lead scientist Philip Shiu's fruit fly whole-brain computational model published in Nature in October 2024. The Shiu team used electron microscopy data of the FlyWire connectome and machine learning-predicted neurotransmitter types to build a leaky integrate-and-fire model covering over 125,000 neurons and 50 million synaptic connections, relying only on the connectivity graph topology, synaptic weights, and excitatory/inhibitory mapping, without the need for tuning parameters neuron by neuron.

The model showed a 91% consistency with 164 published experimental observations. When synaptic weights were randomly shuffled, only 1 out of 100 simulations could reproduce the correct behavior, indicating the strong constraining power of the connectome structure on brain function.

This demonstration integrated the brain model into the NeuroMechFly v2 simulation framework, forming a complete closed loop of sensory input → whole-brain neural activity → motor output. Eon Systems, headquartered in San Francisco, has an advisory board that includes Harvard geneticist George Church and Mathematica founder Stephen Wolfram. The next step is mouse whole-brain simulation (approximately 70 million neurons, 560 times that of a fruit fly), with the ultimate goal being human whole-brain simulation.