Researchers at EPFL, Duke University, and Instituto Superior Técnico have developed a novel approach to understanding how neural circuits work by combining computational neuroscience with robotics. Rather than surgically manipulating living organisms, the team created a physics-based digital zebrafish and a swimming robot powered by biologically inspired algorithms, allowing them to interrogate the relationship between physical bodies and brain function in ways previously impossible.

The research focuses on a fundamental behavior called the optomotor response (OMR), which allows fish to maintain position in flowing water by detecting visual motion and adjusting their swimming accordingly. According to Robohub, this represents one of the clearest examples of how sensory input transforms into motor output through neural pathways.

Building a Virtual Nervous System

The team's breakthrough centers on simZFish, a software model that replicates a six-day-old larval zebrafish at full scale: a 4-millimeter body with simulated eyes, fins, and an artificial neural circuit derived from actual brain-imaging data. The model runs within Webots, a physics engine that simulates water dynamics and physical interactions with precise accuracy.

What sets this approach apart is its experimental flexibility. Researchers can:

  • Disable specific neurons or connections to test their necessity
  • Modify optical properties like lens shape and focal length
  • Alter body segments and fin geometry
  • Observe behavioral consequences instantly

This would be impossible in living animals. Traditional neuroscience can only correlate neural firing patterns with observed behavior, but cannot definitively prove causation or easily test 'what if' scenarios without invasive genetic or surgical intervention.

From Simulation to Hardware

The team validated their digital model by constructing a free-swimming robot embodying the same neural architecture and physical form. This robotic fish successfully navigates upstream using only visual input and the same decision-making circuits found in biological larvae, even under poor visibility conditions.

The success of the robot demonstrates that the simulated neural circuits capture essential algorithmic principles. More importantly, it reveals that neither the brain nor the body alone explains behavior: the fish's morphology, including eye placement and body dynamics, fundamentally shapes how its nervous system solves navigation problems.

Implications for AI and Neuroscience

This convergence of neuroscience, simulation, and robotics points toward new methodologies for understanding biological intelligence. By using AI-driven models grounded in empirical neural data, researchers sidestep ethical concerns about animal experimentation while accelerating hypothesis testing.

The findings suggest that embodied cognition research tools like simZFish could accelerate discovery in neuroscience by enabling rapid iteration on circuit designs. Future applications might include engineering artificial nervous systems for soft robotics or understanding how genetic changes affect neural-behavioral relationships without animal testing.

The work exemplifies a broader shift in computational biology toward digital twins: high-fidelity virtual organisms that serve as testbeds for biological hypotheses before validation in living systems.