In a groundbreaking advancement at the intersection of neuroscience, robotics, and biomechanics, researchers at the BioRobotics Laboratory of the École Polytechnique Fédérale de Lausanne (EPFL) have unveiled an extraordinary new approach to unraveling the complexities of brain-body-environment interactions in vertebrates. Their pioneering work, recently published in the prestigious journal Science Robotics, centers on an artificial larval zebrafish robot, termed Zbot, that integrates real-time neural circuitry with physical embodiment and environmental interaction to model and investigate visuomotor behavior in unprecedented detail.
Understanding the brain’s function has traditionally involved dissecting neural circuits in isolation under highly controlled laboratory conditions. Yet, a persistent paradox in neuroscience is that natural neural activity and behavior arise from a brain embodied within a living organism interacting dynamically with its surroundings. Without incorporating the body and environment, clues about how sensory inputs shape neural computations and motor outputs remain elusive. By designing an artificial organism that combines neural simulations derived from live animals with robotic substrate capable of swimming and perceiving sensory cues in a naturalistic habitat, EPFL’s BioRobotics Lab provides a revolutionary platform to study embodiment—the fundamental principle that the body profoundly influences brain function.
At the core of this research is the larval zebrafish, a diminutive model organism favored in neuroscience due to its translucency and genetic accessibility, allowing optical access to its entire brain. Neurobiologist Eva Naumann at Duke University provided a detailed neural network architecture for this species based on cutting-edge real-time calcium imaging techniques, which record neuronal activity at single-cell resolution while fish respond to visual stimuli. Harnessing these data, Naumann’s team characterized key visuomotor behaviors like the optomotor response—a reflex allowing fish to orient and swim against flowing water currents to maintain station in streams. These findings set the neural benchmark for EPFL’s robotic and simulated models.
The researchers at BioRobotics synthesized a complex simulation integrating visual processing in the retina, neuronal circuitry across brain regions, and the spinal cord’s motor commands, coupled to biomechanically accurate body kinematics of the larval fish. This virtual zebrafish, envisioned as an embodied computational organism, was subjected to simulated water flow and dynamic visual scenes that mimic natural aquatic environments. Remarkably, the computer model replicated the nuanced swimming corrections fish employ to compensate for water displacement and maintain position, demonstrating that the computational design successfully reverse-engineered the sensorimotor circuits underlying larval zebrafish behaviors.
Further analysis within the simulation revealed that the majority of neural signals driving behavioral responses originate from a focused region of the retina, a previously underappreciated insight into the organization of visual inputs critical for orientation. Intriguingly, the researchers’ model predicted the existence of two novel neuron types necessary to explain the behavioral responses elicited by complex and atypical visual stimuli. These computational predictions pave the way for future physiological experiments to validate the existence and function of these elusive neural elements.
To transcend simulations and validate these discoveries in the physical world, EPFL postdoctoral researcher Xiangxiao Liu engineered a striking 80-centimeter robotic larval zebrafish. Named Zbot, this biomimetic robot is outfitted with dual cameras functioning as eyes and sophisticated motors replicating the segmented tail movements characteristic of live zebrafish. Crucially, the same neural control circuits instantiated in computer models were embedded within Zbot’s control system. Deploying Zbot into the dynamic currents of Lausanne’s Chamberonne River allowed the team to observe real-time visuomotor coordination rooted in biologically inspired neural architectures.
In these naturalistic riverine experiments, Zbot consistently demonstrated the optomotor reflex, swimming upstream and maintaining its station despite turbulence and chaotic flow patterns. This embodied manifestation of neural circuitry highlights the critical role of physical instantiation and environmental feedback in brain function. More so, it showcases that even amid the behavioral randomness innate to biological systems, intrinsic circuit dynamics converge robustly to reorient an organism against environmental perturbations—a fundamental survival mechanism.
Beyond the immediate empirical insights, the implications of this research are profound and multi-disciplinary. First, it validates the hypothesis that visual inputs alone are sufficient for locomotor compensation in zebrafish, isolating sensation and motor control mechanisms in ways impossible in vivo due to the entanglement of multiple sensory modalities. Additionally, by sharing their simulation platform and robot designs as open-source resources, the BioRobotics Lab invites the global scientific community to extend these approaches to other species and sensorimotor systems, accelerating discovery across neuroscience, ethology, and robotics.
This work deftly demonstrates the importance of artificial embodied models not only for hypothesis testing but also for uncovering unknown neural components and behaviors. In traditional animal experiments, delineating sufficiency versus necessity of sensorimotor pathways is constrained by biological limitations—one cannot simply “turn off” all pathways except one. Here, however, the controlled environment of the simulation and robotic platform allows researchers to isolate and manipulate variables systematically, gaining insights into the minimal circuits required for behavior.
Furthermore, the integration of biomechanics with neural control in a physical agent poised in a natural habitat marks a transformative step in robotics, where biomimetic design informs both engineering and biological understanding. Unlike classical robotics that operate in simplified or artificial contexts, Zbot embodies the complex, stochastic challenges real organisms face, making it a valuable testbed for evolutionary and comparative studies on sensorimotor adaptation.
With further developments underway at the BioRobotics Lab focused on unraveling the complexities of zebrafish swimming patterns and multisensory coordination, this research sets the stage for a new era of embodied neuroscience. By bridging the molecular and circuit-level insights emerging from neuroscience with the tangible, robotic reenactment of animal behavior, the team not only illuminates fundamental principles of brain function but also inspires innovations in autonomous robots capable of sophisticated, adaptive behaviors in fluid environments.
In conclusion, the artificial embodied circuits pioneered by the EPFL BioRobotics Laboratory reveal how the intimate coupling between brain, body, and environment shapes behavior in vertebrates. Such integrative models underscore the necessity of holistic approaches beyond isolated neural observations and herald transformative potentials for neuroscience, robotics, and beyond. This seminal work, realized through the concerted efforts of interdisciplinary expertise, represents a defining stride toward decoding how nature’s sensorimotor architectures yield robust and flexible animal behaviors in the intricacies of the real world.
Subject of Research: Embodied neural circuits and sensorimotor coordination in larval zebrafish
Article Title: Artificial Embodied Circuits Uncover Neural Architectures of Vertebrate Visuomotor Behaviors
News Publication Date: 15 October 2025
Web References: https://doi.org/10.1126/scirobotics.adv4408
References: Published in Science Robotics, DOI: 10.1126/scirobotics.adv4408
Image Credits: 2025 BioRob EPFL CC BY SA 4.0
