Embodied Neural Circuits in Zebrafish

Journal Cover Feature & International Recognition

This research was featured on the cover of Science Robotics (October 2025, Volume 10, Issue 107) and received international media coverage, including features in Italian national news outlets.

Media Coverage:

• ANSA (https://www.ansa.it/canale_scienza/notizie/frontiere/2025/11/06/da-un-robot-con-il-cervello-di-pesce-un-nuovo-approccio-allia_20cc2205-5ddc-439d-a896-d2f4d6c8c5d9.html)
• Repubblica (https://www.repubblica.it/tecnologia/2025/10/30/news/zebrafisch_scienza_pesce-robot_svela_i_segreti_del_cervello_e_del_movimento-424948228/)
• La Stampa (https://www.lastampa.it/tecnologia/2025/10/30/news/zebrafisch_scienza_pesce-robot_svela_i_segreti_del_cervello_e_del_movimento-424948228/)
• IVG (https://www.ivg.it/2025/11/un-giovane-ricercatore-di-bergeggi-tra-gli-autori-di-uno-studio-pubblicato-su-science-robotics/)
• EPFL Press Release (https://actu.epfl.ch/news/roboticists-reverse-engineer-zebrafish-navigatio-2/)
• Duke University Press Release (https://medschool.duke.edu/news/robotic-fish-unlocks-secrets-brain-body-connection)

The Challenge: Understanding Brains in Isolation

For decades, neuroscience has studied neural circuits by isolating them from the bodies they control and the environments they navigate. But brains didn’t evolve in isolation—they evolved within specific bodies, interacting with physical environments. To truly understand how neural circuits work, we need to study them as part of an integrated brain-body-environment system.

This project tackled that challenge head-on by building a complete neuromechanical simulation of the zebrafish optomotor response—a fundamental behavior where fish stabilize their position by swimming against visual motion. By computationally reproducing everything from neural activity to body mechanics to water hydrodynamics, we could test hypotheses impossible to address in biological experiments alone.

My Role in the Project

I worked at EPFL’s BioRob Laboratory for over 18 months across two semester projects, my master’s thesis, and an additional summer, contributing directly and significantly to this publication. My work centered on three key areas:

Software Development:

• Built the robotic control software for the physical zebrafish robot (ZBot)
• Co-developed the neuromechanical simulation platform (simZFish)
• Created experimental frameworks for systematic testing and data collection

Hardware & Experimental Work:

• Participated in designing and constructing the physical robotic platform
• Conducted extensive experiments in both controlled laboratory settings and real river environments
• Performed validation experiments comparing simulated and real robot behaviors

Data Analysis & Validation:

• Analyzed behavioral data across simulation, robotic, and biological experiments
• Validated simulation predictions against experimental observations
• Contributed to integrating findings from neural imaging experiments back into the model

The Approach: Iterating Between Simulation and Reality

The power of this work came from constantly moving between computational models, biological experiments, and physical robots. We built a realistic neuromechanical simulation that reproduced:

• Neural circuits derived from experimental data on real zebrafish brains
• Body mechanics capturing physical structure and movement
• Hydrodynamics capturing how the fish body interacts with water
• Visual processing modeling the zebrafish eye and retinal connectivity
• Environmental context from simple visual patterns to turbulent river flows

This simulation became a platform for discovery. We could manipulate features impossible to change in living animals—tweaking lens properties, rewiring retinal connections, testing behaviors in virtual environments—and then validate our findings experimentally.

Key Discoveries: When Simulation Predicts Biology

The tight coupling between simulation and experiment led to genuine discoveries:

1. Why Fish Use Their Lower Visual Field

By systematically changing the zebrafish lens properties and retinal connectivity in simulation, we discovered why the lower posterior visual field drives the strongest optomotor responses. The simulation predicted specific receptive field properties that were then confirmed in real zebrafish, explaining a long-standing observation about how these fish process visual motion.

2. Predicting New Neural Response Types

When we challenged the simZFish with novel visual stimuli, it predicted neuronal response patterns that hadn’t been observed before. Working with experimentalists, we then used two-photon calcium imaging to look for these responses in the brains of living zebrafish—and found them. These discoveries were incorporated back into the simulation, completing the loop between computational prediction and biological validation.

3. Autonomous Navigation in Complex Environments

We placed the simZFish in virtual rivers where it had to maintain position against water flow. Without any explicit instructions on how to solve this problem, the embodied sensorimotor circuits learned to use current-induced optic flow patterns as navigational cues, autonomously compensating for the simulated water flow.

Validation with Physical Robots in Real Rivers

Simulations are powerful, but they’re still approximations. To truly validate our findings, we built a physical robot—the ZBot—and tested it in a real river with all the complexity that simulations can’t fully capture: turbulent flows, varying light conditions, unpredictable environmental noise.

The robot successfully maintained its position in the river using the same embodied sensorimotor control principles we’d developed in simulation. This wasn’t just an engineering demonstration—it was proof that the principles we’d discovered about how embodiment shapes neural function actually work in the messy, complex real world.

I was heavily involved in these river experiments, from preparing the robot hardware to running trials in challenging environmental conditions to analyzing the resulting behavioral data.

Impact: A New Paradigm for Neuroscience

This work demonstrates a fundamentally different way of studying neural circuits—one that embraces the complexity of brain-body-environment interactions rather than trying to eliminate it. By building realistic embodied models and iterating between simulation, biological experiments, and physical robots, we can ask questions and test hypotheses that would be impossible with any single approach.

The framework we developed isn’t limited to zebrafish. The same principles—building realistic neuromechanical models, validating with biology, testing with robots—can be applied to understand sensorimotor processing in other systems, from insects to mammals.