Which Animals Can Be Fully Implemented as Behavioral Artificial Intelligence?

As an Amazon Associate we earn from qualifying purchases.

Key Takeaways

• Nematode worms and adult fruit flies now run as connectome-driven digital twins inside virtual physics bodies.
• Deep reinforcement learning networks replicate three-dimensional rat movement kinematics and motor cortex activity.
• Complete behavioral identity requires pairing nanoscale synaptic maps with neuromodulators and embodied physics.

Why Whole-Brain Emulation Separates Structural Wiring From Behavioral Artificial Intelligence

In September 2026, researchers at the Howard Hughes Medical Institute Janelia Research Campus, the University of Cambridge, and Google Research published the complete male fruit fly connectome, charting more than 166,000 neurons and 125 million synaptic connections across the insect brain and ventral nerve cord. That milestone followed the June 2026 release of a complete female central nervous system map by Harvard Medical School and the Princeton-led FlyWire Consortium, alongside March 2026 demonstrations by startup Eon Systems placing a simulated fruit fly brain inside a virtual body. Together, these achievements shifted behavioral artificial intelligence (AI) from theoretical speculation into an empirical engineering discipline. Scientists can now evaluate which living organisms have been, or soon can be, instantiated in software so that the digital animal responds to stimuli and moves through a physical environment identically to its biological counterpart.

Answering which species have crossed this threshold requires distinguishing between two fundamentally different engineering pathways that often merge in public discussion. The first pathway is whole-brain emulation (WBE), which extracts intelligence directly from biological tissue by slicing and imaging an entire nervous system at nanometer resolution, reconstructing every cell and synapse into a wiring diagram known as a connectome, and running biophysical equations that replicate electrochemical signaling. In his foundational book Connectome: How the Brain’s Wiring Makes Us Who We Are, Princeton neuroscientist Sebastian Seung established how nanoscale synaptic mapping provides the structural foundation for testing mechanistic models of identity, memory, and behavior.

The second pathway is biomechanical behavioral cloning, which bypasses nanometer-scale tissue slicing and instead trains deep reinforcement learning networks on high-resolution three-dimensional motion-capture recordings of living animals inside physics engines. Rather than copying each biological synapse one by one, behavioral cloning optimizes an artificial neural network to solve the same inverse dynamics and motor-control problems that a living brain solves when coordinating muscles, tendons, and skeletal joints. As examined in New Space Economy’s overview of digital twinning in the space economy, virtual replicas derive their operational value from coupling a physical object, a mathematical model, and continuous sensor feedback to predict behavior under changing physical forces.

A third hybrid approach interpolates mesoscale cellular atlases and axonal projection databases into large-scale spiking neural networks that approximate whole-brain dynamics without requiring electron microscopy of every individual synapse. Conversely, recent control experiments have shown that coupling a real biological connectome to a foreign animal body through reinforcement learning adapters can produce deceptively lifelike locomotion that carries zero biological validity. Evaluating which animals currently operate as full behavioral AI requires inspecting structural completeness, biophysical accuracy, and closed-loop sensorimotor embodiment across each model organism.

The table below compares the four primary computational paradigms used to construct virtual animal brains and bodies as of October 2026.

Caenorhabditis Elegans and the Four-Decade Quest to Embody 302 Neurons

The first animal ever selected for complete digital implementation was Caenorhabditis elegans, a transparent, millimeter-long soil nematode whose adult hermaphrodite nervous system contains exactly 302 neurons and approximately 7,000 chemical synapses and electrical gap junctions. In 1986, John White and colleagues at the Medical Research Council Laboratory of Molecular Biology published the first full wiring diagram of the nematode nervous system after more than a decade of manual serial-section electron microscopy. Subsequent high-resolution re-mappings in 2019 and 2025 expanded this structural record across both sexes and multiple developmental stages, making the roundworm the longest-standing testbed for whole-organism simulation.

Beginning in 2011, the international OpenWorm project set out to transform that static wiring diagram into a cellular-level digital twin capable of autonomous movement. As documented in the consortium’s foundational open-science architecture paper, engineers combined two distinct software engines: c302, a multi-scale neuroml framework that models the electrical activity of the 302 neurons and 95 body-wall muscle cells, and Sibernetic, a smoothed-particle hydrodynamics solver that simulates the mechanics of the worm’s hydrostatic skeleton swimming through viscous fluid. When sensory neurons in the virtual head receive simulated chemical or mechanical inputs, electrical signals propagate through the connectome and trigger alternating dorsal and ventral muscle contractions that propel the digital nematode forward in characteristic sinusoidal waves.

Researchers also uploaded the nematode connectome into physical wheeled robots and soft-bodied biomimetic machines without writing hand-coded navigation rules. When sonar or infrared proximity sensors on the robot stimulated the digital equivalents of the worm’s nose-touch mechanosensory neurons, signal propagation through the 302-neuron network caused the robot to back away, turn, and resume forward foraging. These demonstrations proved that a biological wiring diagram can close the sensorimotor loop inside synthetic hardware.

Why a Complete Wiring Diagram Alone Does Not Guarantee Identical Worm Behavior

Despite nearly four decades of structural mapping, neuroscientists emphasize that running the static C. elegans connectome in software does not yet reproduce the full behavioral repertoire of a living worm across all timescales. As analyzed in an April 2026 technical assessment of whole-brain emulation and AI, whole-brain emulation decomposes into two separate challenges: structural scanning, which reads out which cells connect to which, and functionalization, which determines the mathematical rules governing how each neuron computes and adapts over time.

Three biological realities explain why a connectome-only nematode simulation still diverges from a real animal in complex laboratory assays:

• Unknown Synaptic Polarities and Conductance Weights: Electron micrographs reveal the physical presence and vesicle count of a synapse, but they do not directly indicate whether a given chemical connection is excitatory or inhibitory, or how strongly a postsynaptic receptor responds in vivo.
• Extrasynaptic Wireless Neuropeptide Signaling: The nematode nervous system relies heavily on volume transmission, in which monoamines such as serotonin and dopamine alongside more than 250 neuropeptides diffuse through extracellular fluid to modulate distant circuits that share no physical synaptic contact.
• Internal Metabolic and Reproductive States: Feeding history, satiety, egg-laying states, and associative learning continuously alter ion-channel expression and synaptic efficacy over minutes to hours, causing an identical anatomical circuit to generate opposite motor responses under different physiological conditions.

Further underscoring the gap between structural wiring and functional performance, an October 2026 connectome reservoir computing study converted published C. elegans connectomes from multiple developmental ages into echo-state recurrent networks and benchmarked them against randomized null networks of equal size. The researchers found that randomized wiring frequently matched or outperformed the raw biological topology when biophysical parameters were held uniform, confirming that anatomical connectivity must be paired with cell-specific physiological parameters to replicate natural behavior.

How Drosophila Melanogaster Achieved Closed-Loop Whole-Brain Emulation

Whereas the 302-neuron roundworm revealed the subtleties of biochemical modulation, the common fruit fly (Drosophila melanogaster) became the centerpiece of whole-brain behavioral AI between 2024 and 2026. Representing more than a 400-fold increase in neuron count and a 7,000-fold increase in synaptic complexity over C. elegans, the adult fruit fly executes rich behaviors including visual landmark navigation, aerial obstacle avoidance, olfactory associative learning, courtship singing, rival aggression, grooming, and feeding.

The breakthrough began in October 2024, when the Princeton-led FlyWire Consortium published the first complete adult brain connectome of a female fruit fly in Nature. By slicing a single fly brain into more than 7,000 nanometer-thin sections, imaging them with high-throughput transmission electron microscopy, applying machine learning segmentation, and coordinating thousands of human proofreaders, the consortium mapped 139,255 neurons and approximately 50 million chemical synapses across 8,453 annotated cell types. In a companion Nature study published the same day, Philip Shiu and colleagues constructed a computational leaky integrate-and-fire model of the entire FlyWire connectome. Assigning excitatory or inhibitory neurotransmitter identities to each neuron, the team stimulated virtual sugar-sensing and water-sensing gustatory receptors and predicted real biological motor neuron activations governing feeding and antennal grooming with more than 90 percent accuracy (reaching 95 percent concordance across validated sensorimotor pathways).

Bridging the Brain and Ventral Nerve Cord to Drive a Virtual Fly Body

A disembodied brain model can predict which descending neurons fire in response to a taste or visual cue, but it cannot walk, groom its wings, or drink from a bowl without a spinal-cord equivalent and a physical body. Two major advances in 2026 closed that sensorimotor loop. First, in March 2026, Eon Systems integrated the conductance-based FlyWire brain simulation with NeuroMechFly v2, a biomechanically accurate virtual fruit fly body developed by Pavan Ramdya’s laboratory at the Swiss Federal Institute of Technology Lausanne (EPFL) and documented in the EPFL NeuroMechFly v2 study. Running inside the MuJoCo physics simulator at one-millisecond time steps on distributed graphics processing units, the embodied fruit fly simulation received virtual sensory inputs, propagated spiking activity across 50 million synapses, and drove 42 leg and body degrees of freedom to stretch its limbs, rub its front tarsi together in grooming sequences, and extend its proboscis to drink.

Second, in June 2026, researchers at Harvard Medical School and Princeton University published the complete Brain and Nerve Cord (BANC) connectome of a female fruit fly, encompassing roughly 160,000 neurons across the entire central nervous system. Three months later, in September 2026, HHMI Janelia Research Campus, Google Research, and the University of Cambridge released the complete male central nervous system connectome in Cell, mapping more than 166,000 neurons and 125 million synapses across the central brain, optic lobes, and ventral nerve cord. Analysis of the full central nervous system revealed a foundational principle for behavioral AI: fruit fly motor control is highly distributed. Rather than relying on a single centralized executive controller in the head, local neural modules within the ventral nerve cord independently coordinate individual leg joints and wing hinges, communicating laterally between limbs to generate stable tripod walking gaits.

The table below summarizes the nervous system scale and current digital implementation status of the five primary animal models in behavioral AI research as of October 2026.

Biomechanical Digital Twins and the Harvard-DeepMind Virtual Rodent

Scaling nanometer-resolution electron microscopy from an insect brain to a mammalian brain requires crossing a vast volumetric divide. An adult brown rat (Rattus norvegicus) possesses approximately 200 million neurons, and a house mouse (Mus musculus) contains roughly 70 million to 71 million neurons and 100 billion synapses, or about 550 times the neuronal count of a fruit fly. Because complete nanometer-scale connectomes do not yet exist for any mammal, neuroscientists and AI engineers developed a reverse-engineering strategy that starts from outward biomechanical behavior and works inward toward neural control.

In June 2024, neurobiologist Bence Ölveczky, graduate researcher Diego Aldarondo, and colleagues at Harvard University partnered with Google DeepMind to publish a landmark virtual rodent neural activity study in Nature. As detailed in the Harvard Gazette report, the team recorded high-resolution three-dimensional motion-capture data from freely moving rats performing natural foraging, rearing, running, and obstacle-leaping behaviors. They then built a biomechanically realistic skeletal and muscular rat body inside the MuJoCo physics engine, complete with gravity, joint torque limits, ground reaction forces, and virtual proprioceptive and visual sensors.

Using deep reinforcement learning, the researchers trained an artificial neural network to serve as the virtual rodent’s brain. Crucially, the network was trained to implement inverse dynamics models: given a reference three-dimensional movement trajectory observed in a real rat, the artificial network had to calculate and output the exact muscular forces required to drive the simulated skeleton along that path in real time. Once trained, the virtual rat generalized to novel movement sequences not included in its training set. When the Harvard team compared internal activations of the artificial neural network against single-neuron electrophysiological recordings from the sensorimotor striatum and motor cortex of real rats performing the same movements, the artificial network accurately predicted the structure and timing of biological neural activity.

This biomechanical digital twin approach parallels engineering practices across aerospace and terrestrial robotics. In New Space Economy’s analysis of Komatsu’s lunar construction blueprint, engineers use physics-based digital twins to test articulated multi-crawler locomotion across uneven lunar craters before fabricating physical prototypes. Similarly, the Harvard-DeepMind virtual rodent provides a transparent computational laboratory in which researchers can lesion specific artificial network layers to study how motor diseases such as Parkinson’s disrupt limb coordination, and roboticists can transfer biological inverse-dynamics policies into agile quadrupedal machines.

Larval Zebrafish and Mouse Digital Twin Brains on the Vertebrate Frontier

Between the 166,000-neuron fruit fly and the 200-million-neuron rat sit two vertebrate model organisms that bridge structural connectomics and whole-brain functional modeling: the larval zebrafish (Danio rerio) and the house mouse (Mus musculus). Each species addresses a distinct bottleneck on the roadmap toward vertebrate whole-brain emulation.

Transparent Larval Zebrafish and the Fish Fire and Wire Dataset

The larval zebrafish is one of the very few vertebrates whose entire brain is small enough, containing approximately 100,000 neurons and 100 million synapses, to be imaged from end to end at cellular resolution using contemporary electron microscopy. Equally important, the zebrafish body and skull remain optically transparent during its larval stage. That transparency allows neuroscientists to express genetically encoded calcium or voltage indicators in every neuron and record live whole-brain electrical activity with light-sheet microscopy as the fish responds to visual motion flows, hunts paramecia, or executes tail-flip escape maneuvers.

As Google Research and Harvard researchers highlighted in their September 2026 connectomics update and in the Frontiers in Neuroscience connectome review, the Fish Fire&Wire and ZAPBench initiatives have combined whole-brain functional calcium recordings with nanometer-scale serial-section electron microscopy in the exact same larval zebrafish specimen. By observing every neuron firing in a living vertebrate and then reconstructing the exact synaptic wiring of that same individual brain, scientists can directly solve the functionalization problem that hampered earlier C. elegans models. Rather than guessing synaptic weights from static anatomy alone, machine learning optimizers can fit biophysical neuron models directly to measured input-output spike trains across all 100,000 vertebrate neurons.

MouseDTB and Mesoscale Spiking Models of the Mammalian Brain

For the adult mouse brain, complete nanometer-resolution electron microscopy remains restricted to cubic-millimeter cortical fragments, which represent roughly one five-hundredth of total mouse brain volume. To study whole-brain mouse behavior before complete nanoscale scanning becomes feasible, computational neuroscientists developed MouseDTB, a single-neuron-resolution Mouse Digital Twin Brain presented in November 2025.

Instead of waiting for a petabyte-scale electron microscope scan of all 70 million mouse neurons, the MouseDTB pipeline integrates high-resolution mesoscale axonal projection maps and spatial cell-type distributions from the Allen Mouse Brain Cell Atlas. The model instantiates tens of millions of leaky integrate-and-fire spiking neurons with conductance-based synapses and uses diffusion ensemble Kalman filtering and hierarchical Bayesian inference to fit regional connectivity parameters against empirical neuroimaging. In validation benchmarks, MouseDTB accurately reproduced resting-state blood-oxygen-level-dependent (BOLD) signals and generated correct behavioral decisions during simulated olfactory Go/No-Go odor discrimination tasks. Although MouseDTB is a statistical cell-atlas twin rather than an individual synapse-by-synapse upload, it demonstrates that whole-brain spiking architectures can already link sensory perception to behavioral choice at mammalian scale.

Why Behavioral Mimicry Can Diverge From Biological Reality

A central methodological question in behavioral AI is whether an artificial animal that walks, grooms, or forages identically to a real animal actually computes in the same way as the biological organism. In Grace Lindsay’s book Models of the Mind: How Physics, Engineering and Mathematics Have Shaped Our Understanding of the Brain and Matthew Cobb’s historical study The Idea of the Brain: The Past and Future of Neuroscience, both authors warn that mathematical models can reproduce outward behavioral phenomena yet rely on internal mechanisms that differ sharply from living tissue.

In March 2026, researchers provided a striking experimental demonstration of this pitfall in a March 2026 digital sphinx study titled The Digital Sphinx: Can a Worm Brain Control a Fly Body?. To test whether behavioral realism proves neurobiological validity, the authors constructed a virtual interspecies chimera combining two phylogenetically distant animals: the 302-neuron connectome of the C. elegans nematode worm and the multi-jointed biomechanical body of the adult Drosophila fruit fly. Sensory feedback from the fly’s legs was fed into the worm’s sensory neurons, and a deep reinforcement learning readout layer was trained to map the worm’s motor neuron activations onto the fly’s leg actuators.

Remarkably, the resulting “digital sphinx” produced highly realistic, coordinated six-legged fruit fly walking. Yet the internal neural circuit driving that locomotion belonged to a legless microscopic worm and taught researchers nothing about how either C. elegans or Drosophila actually controls movement in nature. Because deep reinforcement learning is a powerful function approximator, an adapter network can exploit almost any recurrent dynamical reservoir, even the connectome of an unrelated species, to generate lifelike limb kinematics.

This finding carries major implications for both neuroscience and AI engineering. As discussed in New Space Economy’s analysis of explainable AI in space, external task accuracy does not guarantee internal mechanistic interpretability. For a virtual animal to serve as a true biological digital twin capable of predicting how genetic mutations, pharmacological compounds, or neural injuries alter behavior, every component of the simulation, from receptor kinetics and synaptic weights to muscle attachment geometry, must be constrained by biological measurements rather than patched together by unconstrained reinforcement learning weights.

The table below details the four primary technical bottlenecks that separate current virtual animals from complete, long-horizon biological identity.

Applications Across Autonomous Space Robotics, Bioastronautics, and Ethics

The ability to implement whole animal nervous systems and biomechanical digital twins in software extends far beyond academic neurobiology, intersecting directly with autonomous robotics, orbital biomedical research, and AI governance. Biological nervous systems evolved under severe mass, volume, and energy constraints that mirror the engineering limits of spacecraft and planetary rovers. An adult fruit fly brain executes real-time visual odometry, collision avoidance, odor plume tracking, and six-legged terrain adaptation on roughly 10 microwatts of metabolic power, whereas conventional graphics processing clusters consume thousands of watts to run deep learning vision models.

As detailed in New Space Economy’s review of the UK’s AI roadmap for space and its analysis of architecting the interplanetary supply chain, deep-space exploration and in-orbit servicing require resource-efficient edge computing and hybrid physics-informed autonomy. Translating the distributed ventral nerve cord modules discovered in the 2026 Drosophila BANC connectome into neuromorphic hardware allows legged planetary explorers to coordinate limb movement locally at each joint without routing every motor command through a power-hungry central processor. In A Thousand Brains: A New Theory of Intelligence, neuroscientist Jeff Hawkins similarly emphasizes how distributed reference frames and sensorimotor integration enable resilient navigation in unfamiliar environments.

In orbital biomedicine, every animal species at the forefront of whole-brain emulation, namely C. elegans, Drosophila melanogaster, larval zebrafish, and rodents, also serves as a primary model organism aboard the International Space Station (ISS) and upcoming commercial stations. As examined in New Space Economy’s comprehensive analysis of ISS scientific discovery and its profile of Starlab’s AI-powered digital twin station, space agencies study nematodes, fruit flies, and rodents in weightlessness to measure how cosmic radiation and altered gravity affect neural adaptation, gene expression, and muscle atrophy. Pairing physical spaceflight habitats with validated computational digital twins of C. elegans, Drosophila, and rodents, alongside CRISPR gene-editing experiments in space, allows bioastronautics teams to run thousands of virtual microgravity and radiation perturbation trials in silico before launching scarce biological payloads into orbit.

Finally, scaling whole-brain emulation from insects toward zebrafish and mammals raises foundational questions in machine welfare and philosophy of mind. Debates examined in New Space Economy’s analyses of why software may never become sentient and when sentient AI warrants legal rights center on whether silicon simulations instantiate valenced experience or merely execute formal syntax without cellular metabolism. Whereas adult fruit flies and nematodes show minimal evidence of complex subjective awareness, extending neuron-by-neuron emulation to vertebrate zebrafish and mammalian brains will require pre-deployment welfare audits to ensure that high-speed virtual training environments do not inadvertently instantiate digital pain or aversive distress.

Summary

As of October 2026, two animals have had their complete nervous systems mapped at nanometer resolution and embodied in closed-loop virtual physics environments: the 302-neuron roundworm Caenorhabditis elegans and the 125,000-to-166,000-neuron adult fruit fly Drosophila melanogaster. Through the OpenWorm project, the FlyWire and BANC connectome consortia, EPFL’s NeuroMechFly v2 framework, and Eon Systems’ conductance-based simulations, virtual worms and flies now process simulated sensory inputs and generate coordinated locomotion, feeding, and grooming behaviors. Alongside these connectome uploads, Harvard University and Google DeepMind’s Virtual Rodent has demonstrated that deep reinforcement learning networks trained on three-dimensional motion capture can replicate the complex musculoskeletal kinematics and motor-cortex firing patterns of a living brown rat inside the MuJoCo physics simulator.

At the vertebrate whole-brain frontier, the 100,000-neuron larval zebrafish has become the first vertebrate with combined whole-brain electrical activity recordings and nanoscale structural connectomics from the same individual specimen, positioning it as the next organism likely to achieve end-to-end neuron-level emulation. For the 70-million-neuron mouse brain, mesoscale spiking models such as MouseDTB already reproduce resting-state dynamics and odor-discrimination tasks as nanometer scanning scales beyond cubic-millimeter fragments. Achieving complete, long-horizon behavioral identity across these species requires integrating static synaptic wiring with wireless neuropeptide diffusion, dendritic computation, synaptic plasticity, and biologically grounded body physics rather than relying on unconstrained reinforcement learning adapters.

Appendix: Useful Books Available on Amazon

• Connectome: How the Brain’s Wiring Makes Us Who We Are
• Models of the Mind: How Physics, Engineering and Mathematics Have Shaped Our Understanding of the Brain
• The Idea of the Brain: The Past and Future of Neuroscience
• A Thousand Brains: A New Theory of Intelligence
• Rhythms of the Brain
• How to Create a Mind: The Secret of Human Thought Revealed
• Superintelligence: Paths, Dangers, Strategies
• Rebooting AI: Building Artificial Intelligence We Can Trust
• Human Compatible: Artificial Intelligence and the Problem of Control
• Life 3.0: Being Human in the Age of Artificial Intelligence

Appendix: Top Questions Answered in This Article

Which Animals Have Complete Nervous System Connectomes Mapped as of October 2026?

As of October 2026, complete nanometer-resolution connectomes spanning the entire nervous system have been published for the nematode worm Caenorhabditis elegans (302 neurons) and both female and male adult fruit flies of the species Drosophila melanogaster (approximately 130,000 to 166,000 neurons). Comprehensive connectomes also exist for several smaller invertebrate larvae, and whole-brain cellular datasets combining neural structure and electrical activity have been assembled for the transparent larval zebrafish (Danio rerio), which contains roughly 100,000 neurons.

What Is the Difference Between Whole-Brain Emulation and Biomechanical Behavioral Cloning?

Whole-brain emulation reconstructs an animal’s actual biological nervous system neuron by neuron and synapse by synapse from electron microscopy scans, and then simulates electrochemical signal propagation across that exact wiring diagram. Biomechanical behavioral cloning, by contrast, trains an artificial deep reinforcement learning network on three-dimensional motion-capture data so that a physics-simulated skeleton moves like the real animal without copying every individual biological synapse. Both approaches can generate realistic animal behavior inside virtual physics environments.

Has the Fruit Fly Brain Been Connected to a Virtual Body in Simulation?

Yes. Building on the FlyWire adult brain connectome and computational feeding and grooming models published in Nature in October 2024, researchers at Eon Systems integrated a conductance-based Drosophila brain simulation with EPFL’s NeuroMechFly v2 physics body in early 2026. In that closed-loop simulation, virtual sensory inputs trigger spiking activity across 50 million synapses, and the resulting motor commands drive the virtual fly body to walk, groom its front legs, and drink from a simulated bowl.

Why Does a Complete C. Elegans Connectome Not Automatically Reproduce All Worm Behaviors?

Although the 302 neurons and roughly 7,000 synapses of C. elegans have been mapped since 1986, a static wiring diagram does not capture wireless extrasynaptic signaling by neuropeptides and monoamines that diffuse through extracellular fluid. Furthermore, electron microscopy alone does not specify the exact physiological conductance, ion-channel kinetics, or short-term plasticity rules of every synapse. Without these biochemical and dynamic parameters, connectome-only simulations reproduce basic swimming and crawling but miss state-dependent learning behaviors.

How Does the Harvard and Google DeepMind Virtual Rodent Replicate Real Rat Behavior?

Researchers at Harvard University and Google DeepMind recorded high-resolution three-dimensional movement trajectories from real rats and trained a deep reinforcement learning network to control a biomechanically realistic rat model inside the MuJoCo physics engine. The artificial network learned inverse dynamics models that calculate the precise muscle forces needed to execute agile foraging, rearing, and jumping movements. Internal activations within the trained virtual network accurately predicted real neural activity recorded from the motor cortex and striatum of living rats.

What Did the 2026 Fruit Fly Brain and Nerve Cord Connectomes Reveal About Animal Movement?

The complete female Brain and Nerve Cord (BANC) connectome published in June 2026 and the complete male central nervous system connectome published in September 2026 mapped how the fruit fly brain links to its spinal-cord equivalent. Researchers discovered that complex behaviors such as walking and grooming are not dictated step by step by a single centralized controller in the brain. Instead, distributed local neural circuits within the ventral nerve cord independently control individual legs and wings and coordinate directly with neighboring limb modules.

What Is the Digital Sphinx Experiment and Why Does It Matter for AI Animal Models?

In a March 2026 study, researchers created a virtual chimera called the digital sphinx by connecting the 302-neuron connectome of a C. elegans roundworm to the six-legged biomechanical body of a fruit fly using a deep reinforcement learning adapter. Even though a nematode worm has no legs, the hybrid system produced highly realistic fruit fly walking. The experiment proved that deep reinforcement learning can force an unrelated circuit to mimic outward behavior, showing why true biological emulation requires strict anatomical and physiological grounding.

Why Is the Larval Zebrafish the Leading Candidate for the First Vertebrate Whole-Brain Emulation?

The larval zebrafish possesses a complete vertebrate brain architecture with approximately 100,000 neurons and 100 million synapses, making it small enough to reconstruct end to end with serial-section electron microscopy. Because the larval fish is optically transparent, scientists in the Fish Fire&Wire and ZAPBench projects can record whole-brain calcium and electrical activity in a living specimen before slicing and mapping the exact same brain. Combining functional recordings with structural wiring allows researchers to fit accurate biophysical parameters for every neuron.

How Close Is Neuroscience to Emulating a Complete Mammalian Mouse Brain?

An adult house mouse brain contains approximately 70 million to 71 million neurons and 100 billion synapses, which is roughly 550 times larger in neuron count than a fruit fly brain. Complete nanometer-scale electron microscopy currently covers only cubic-millimeter cortical fragments of the mouse brain. However, data-driven whole-brain spiking models such as MouseDTB, introduced in late 2025, already use mesoscale axonal projection atlases to simulate tens of millions of leaky integrate-and-fire neurons that match mouse resting-state activity and odor-discrimination behavior.

How Do Emulated Animal Brains Support Space Exploration and Orbital Robotics?

Insect and nematode nervous systems achieve agile navigation, multisensory integration, and fault-tolerant locomotion using microwatts of power, providing blueprints for low-power neuromorphic processors aboard planetary rovers and orbital servicing robots. In addition, C. elegans, Drosophila, zebrafish, and rodents are core model organisms used aboard the International Space Station to study microgravity and cosmic radiation. Digital twins of these animals allow bioastronautics researchers to simulate physiological adaptations and screen genetic countermeasures on Earth prior to orbital flight.

Appendix: Glossary of Key Terms

Whole-Brain Emulation

A computational neuroscience and engineering methodology that replicates the complete structure and electrochemical function of a biological brain at the resolution of individual neurons and synapses, aiming to produce a simulation that performs the exact same information-processing computations and sensorimotor transformations as the original living organism.

Connectome

A comprehensive structural wiring diagram of an organism’s nervous system that maps the anatomical locations, dendritic and axonal branches, and chemical or electrical synaptic connections of all neurons within a specific brain region or across the entire central and peripheral nervous system at cellular or nanoscale resolution.

Conductance-Based Neuron Model

A biophysical mathematical representation of a nerve cell that calculates membrane voltage dynamics by solving differential equations for individual ion channels, such as sodium, potassium, and calcium conductances, providing higher physiological fidelity than simplified point-neuron switches at the cost of substantially greater computational processing demand.

Leaky Integrate-and-Fire Model

A computationally efficient spiking neuron model that treats a nerve cell as an electrical capacitor with a leaky resistor, summing incoming excitatory and inhibitory synaptic currents over time until the membrane voltage crosses a fixed threshold, emits an action potential spike, and resets to a baseline resting potential.

Ventral Nerve Cord

The primary trunk of the central nervous system in invertebrates such as arthropods and nematodes, running longitudinally along the belly side of the body and functioning analogously to the vertebrate spinal cord by relaying sensory feedback to the brain and coordinating local motor circuits for limbs and wings.

Inverse Dynamics

A biomechanical and control-theory calculation that determines the exact muscular forces and joint torques required to move a physical or simulated skeleton along a desired three-dimensional trajectory and accounting for gravity, limb inertia, ground reaction forces, and mechanical constraints across interconnected body segments.

Deep Reinforcement Learning

A machine learning methodology in which an artificial neural network agent interacts with a physical or simulated environment through trial and error, adjusting its internal connection weights to maximize cumulative numerical reward signals across complex multi-step tasks such as locomotion, manipulation, or navigation.

Proprioception

The internal sensory modality through which an animal or embodied robotic system continuously perceives the position, joint angles, velocity, mechanical load, and tension of its own limbs, muscles, and body segments, enabling closed-loop posture stabilization and coordinated movement across uneven or shifting terrain.

Neuromodulation

The physiological process in which chemical messengers such as serotonin, dopamine, octopamine, and neuropeptides diffuse across extracellular space to alter the excitability, firing thresholds, and synaptic strengths of broad neuronal populations over seconds to hours, reconfiguring how an anatomically fixed circuit responds to sensory inputs.

Reservoir Computing

A recurrent neural network framework in which inputs drive a fixed, high-dimensional dynamical system called a reservoir, and only a lightweight linear readout layer is trained to map the reservoir’s internal state trajectories onto desired behavioral or computational outputs.