Robots and simulations in the loop: Embodied testbeds for understanding sensory perception

Perception is not a passive readout of the senses. The brain must encode, process, and transform a constant stream of sensory information into stable, actionable percepts, and it does so while the body is moving through the world. This holds across species: the mechanisms that turn sensation into perception have been shaped by the demands of behaving bodies, whether human, mammalian, avian, or invertebrate. Uncovering the neural and behavioural mechanisms underlying this transformation remains one of the central goals of sensory-perceptual neuroscience. Machine learning and deep learning have changed how we model and analyse perceptual and neural data, yet translating insights from artificial networks back into biological sensory mechanisms is rarely straightforward: the models are powerful but disembodied, and their “black-box” behaviour is difficult to map onto how real brains see, hear, touch, and adapt. What is often missing is a body, a system that must perceive, move, and cope with the consequences of its own actions in real time, reflecting the conditions under which perception evolved and operates.This Research Topic puts robots and simulated embodied agents in the loop as testbeds for theories of sensory perception. Physical robots, neurorobotic systems, and closed-loop hardware offer a distinctive means of testing models of sensory encoding and perception in dynamic, interactive settings. High-fidelity simulation offers the same closed sensorimotor loop without the technical barrier of a physical platform, and is already established practice across several sensory domains: neuromechanical models that couple neural control to body dynamics, and auditory agents trained on sound carrying the filtering effects of head and ear anatomy, are two examples. Both approaches are equally in scope. What matters is that the agent closes the loop: its perception shapes its actions, its actions change its sensory input, and the resulting self-generated data reflect the ecologically valid, naturalistic conditions in which perception operates. Robotics and embodied simulation therefore provide not only an application of perceptual theory, but also an experimental proving ground where hypotheses concerning sensory coding, multisensory integration, perceptual decision-making, and adaptation can be embodied, stressed, and falsified.The primary goal of this Research Topic is to elucidate the neural and behavioural mechanisms through which the brain encodes, processes, and transforms sensory information into perception, by combining the analytical power of machine learning with the experimental validation capabilities of embodied platforms, whether physical or simulated. Work on any animal species is welcome, including humans. Human perception remains a central focus, and studies grounded in other species are equally in scope: invertebrate and vertebrate sensory systems have furnished many of the models that embodied agents now instantiate, and comparison across species is frequently where a mechanism becomes visible. We aim to bring together the modelling and interpretation of perceptual and neural data with embodied systems that instantiate and test those models under realistic sensorimotor demands across visual, auditory, somatosensory, olfactory, and gustatory modalities. Machine-learning tools, including modern data-driven and generative approaches, can be used to build and interpret these models, while robots and simulated agents provide the testbeds that ground them in perception and behaviour. By integrating these fields, we aim to reveal how biological brains achieve perception, multisensory integration, perceptual learning, and adaptation, while advancing mechanistic frameworks such as embodied cognition, predictive coding, and active inference, with clear relevance to sensory training, neurorehabilitation, and assistive technologies.We welcome contributions that combine robotics, embodied simulation, and machine learning to advance understanding of sensory perception and its neural basis. We particularly encourage novel computational models, experimental paradigms, and analytical techniques involving embodied agents. Themes may include, but are not limited to:Embodied models of sensory coding validated on robotic or simulated platforms: Instantiating circuit- or systems-level models of perception in physical robots or high-fidelity simulated agents and testing their behavioural predictions.Neuromechanical and sensory simulation: Coupling neural control models to simulated body dynamics or simulated sensory environments to study encoding under realistic physical constraints.Closed-loop sensorimotor experimentation: Using robots or simulated agents to investigate how perception and action shape one another, and to test theories that become experimentally accessible only when the sensorimotor loop is closed.Multisensory integration in embodied agents: Examining how signals across sensory modalities are combined, weighted, and recalibrated, and what these processes reveal about biological perceptual integration.Comparative and cross-species perspectives: Embodied or simulated work grounded in human or non-human sensory systems, and comparisons across species that expose shared or divergent mechanisms.Bio-inspired and neuromorphic sensory processing: Developing visual, auditory, tactile, and other sensory architectures, including spiking and event-driven systems, derived from perceptual principles and evaluated through embodied tasks.Machine learning for interpreting perceptual and neural activity in closed-loop settings: Applying analytical and generative methods to data produced during embodied, interactive perception.Robots and simulated agents as testbeds for theories of perception: Embodied evaluations of frameworks such as predictive coding, active inference, and embodied cognition.Perceptual learning, adaptation, and plasticity: Computational accounts of how brains recalibrate perception, validated in embodied agents interacting with their environments. Topic editors: Kiki Van Der Heijden, Alexandre Bernardino, Xiangxiao Liu, Shunsuke Shigaki Source: https://www.frontiersin.org/research-topics/86816/robots-and-simulations-in-the-loop-embodied-testbeds-for-understanding-sensory-perception