Desert ants are particularly excellent navigators, able to use visual memories to learn long foraging routes and return to their nest in complex natural environments. The mushroom body (MB) is believed to be the key brain region that forms these visual memories. Specifically, the MBs can store two types of parallel and opposing valence memories, treated as attractive and repulsive, respectively. However, the underlying neural computations for integrating two opposing visual memories remain unclear. Based on biological findings, we propose two possible integration mechanisms for these opposing visual memories and implement both in our model. The first method, Subtraction-Based Integration, calculates the familiarity of the view in the current direction by computing the difference in the firing rates of two mushroom body output neurons (MBONs), which then guides directional decisions. The second method, Vector-Sum Integration, computes a preferred direction for each MBON, with the final movement direction determined by summing these directional vectors. We validated the effectiveness and testability of both integration methods by replicating recent biological experiments. Results demonstrates that both models could successfully replicate real ant’s trap-avoiding behavior, offering insights into the mechanisms by which insects could combine appetitive and aversive memories to make appropriate navigational decision.
Animals can accomplish many incredible behavioral feats across a wide range of operational environments and scales that current robots struggle to match. One explanation for this performance gap is the extraordinary properties of the biological materials that comprise animals, such as muscle tissue. Using living muscle tissue as an actuator can endow robotic systems with highly desirable properties such as self-healing, compliance, and biocompatibility. Unlike traditional soft robotic actuators, living muscle biohybrid actuators exhibit unique adaptability, growing stronger with use. While muscle adaptability is a benefit to muscular organisms, it currently presents a challenge for biohybrid researchers: how does one design and control a robot whose actuators’ force output changes over time? Here, we incorporate muscle adaptability into a many-muscle biohybrid robot design and modeling tool, leveraging reinforcement learning as both a co-design partner and system controller. Our results show that adaptive agents outperform non-adaptive agents in terms of maximum rewards and training time. Together, these contributions can enable the elucidation of muscle actuator adaptation and inform the design and modeling of adaptive many-muscle robots.
Recent advancements in rapid prototyping technologies, such as 3D printing, have transformed industries by enabling fast and customizable fabrication. Meanwhile, the ancient art of origami—where complex three-dimensional structures emerge from folding flat sheets—offers a unique, reversible, and geometrically efficient alternative, increasingly explored in fields like robotics and deployable systems. While self-folding origami has shown promise in micro-scale applications, challenges such as torque limitations and hinge friction impede its scalability. To explore how programmable origami mechanisms can be physically realized and publicly experienced at scale, this study presents a large-scale transforming kinetic sculpture designed for an interactive exhibition context. The system employs a tendon-driven actuation mechanism and is fabricated from laser-cut acrylic slices and double-layer PET living hinges, enabling smooth expansion from 0.7 m to 3.5 m in height. The sculpture features both an automated display section and a manually operated audience interaction section, encouraging tactile engagement with structural transformation. During a week-long public exhibition, the installation operated continuously for over 7 h per day and completed an average of 46 actuation cycles daily without failure. This work demonstrates the mechanical robustness and scalability of tendon-actuated origami structures and highlights the potential of kinetic origami as an accessible and programmable medium for public engagement, bridging technical innovation with artistic expression.
This paper presents the design and development of a biologically inspired rat robot. The robot is scaled 2.5 times the size of a female Sprague-Dawley rat. The hindlimbs are each equipped with 4 motors to control the hip, knee, and ankle rotation in the sagittal plane, and internal/external hip rotation. The forelimbs are equipped with five motors to control the scapula, shoulder, elbow, and wrist rotation in the sagittal plane, as well as abduction/adduction of the scapula. Additionally, the hands and feet of the robot are comprised of two sections, connected with a pin and torsional springs. This allows the feet to have passive compliance and conform to the ground while walking. The leg segments are based on scanned rat bones with shapes modified for ease of assembly and 3D printing. Parts are printed using PLA with internal supports for structural reinforcement. The scapula, shoulder, and hip joints are directly driven by motors. The lower joints are driven by motors using pulley-belt transmission systems to allow the motors to be mounted more proximal, reducing the legs’ inertia. This robot was developed as a physical platform to test and make predictions about how the animal and its nervous system may interact with the environment in a more realistic way than in software simulations alone.
Swallowable robots offer a promising frontier for biomedical applications such as targeted drug delivery or sampling within the gastrointestinal tract. This work presents a compact, magnetically actuated crank-slider mechanism fully embedded in a size 000 pharmaceutical capsule (length:25.8 mm, diamater:9.9 mm) that is capable of (1) targeted delivery of substances (sucrose) and (2) mechanical sampling or injection via controlled needle deployment. Both actions are validated through proof-of-concept tests. The device features a permanent magnet attached to a 3D-printed crank-slider system, actuated externally by four electromagnetic coils. This magnetic control strategy allows for the development of compact, electronics-free, and more biocompatible robotic systems. This study presents a fully realized, low-cost, functional prototype that can be scaled and adapted for different applications. The system acts as a programmable, wireless mini-actuator operating without onboard power. Its simplicity, modularity, and remote activation make it relevant to biomimetic devices, and intelligent prosthetic tools for biomedical use.
Backbones are a critical component of bioinspired snake robots, providing support and actuator attachment points. Current soft snake robots overwhelmingly rely on continuum backbones, which are continuous strips or rods. Continuum backbones are straightforward to manufacture, but tie together torsional and bending stiffness, buckling load and maximum curvature. For example, increasing cross-section size to raise buckling load increases torsional stiffness and bending stiffness in at least one plane. Biological snakes have highly articulated backbones that break these scalings, but they also have complex vertebra geometry and musculature. In this work, we develop a concept for a highly articulated, snake-inspired backbone and a corresponding actuator layout. The artificial vertebrae include key features identified in biology literature, such as articulation points and motion limiters. We measure range of motion in two planes, note emergent twist, and demonstrate locomotion. The results show feasibility of an alternative soft snake robot design, with closer mimicry of biology and significantly higher mechanical complexity. Emergent behaviors tying twist and bending deformation suggest a path forward for producing complex movements from simple actuation inputs, but further work is needed to model robot mechanics.
Measuring awareness in artificial agents remains an unresolved challenge. We argue that it holds untapped potential for enhancing their design, control, and effectiveness. In this paper, we propose a novel and tractable approach to measure the impact of awareness on system performance, structured around distinct dimensions of awareness – temporal, spatial, metacognitive, self and agentive. Each dimension is linked to specific capacities and tasks. Specifically, we demonstrate our approach through a swarm robotics intralogistics scenario, where we assess the influence of two dimensions of awareness – spatial and self – on the performance of the swarm in a collective transport task. Our results reveal how increased abilities along these awareness dimensions affect overall swarm efficiency. This framework represents an initial step towards quantifying awareness in, and across, artificial systems.
Many insects, such as ants, are highly adept navigators, using learned visual information for route navigation and homing. They do this despite low resolution vision and small brains. Whether insects are also capable of more complex spatial cognition, such as pose independent place recognition, is an open question. In this study we first explored whether Convolutional Neural Networks (CNNs) of varying size are capable of a real world ‘ant’s eye’ place recognition task. We collected panoramic images from a set of 11 distinct places with variations in pose, time of day, weather and season. The CNNs were trained to categorise the places for input images of varying resolution. Whilst VGG16 performed best for all image resolutions, there was a general trend relating model size and image resolution to performance. Of the custom models, smaller models learn lower resolutions better than higher resolutions, and visa versa. We also found that resolutions 113 × 36 and 57 × 18 elicit the first or second best performance of the custom models, suggesting optimal performance for low computational processing lies between these two highlighted resolutions.
Braided pneumatic actuators (BPAs) have become a useful resource in the field of biomimetic robotics. Despite being difficult to control, the force-length and force-velocity characteristics present in this soft actuator offer an opportunity for a more biologically-oriented form of control. In addition, BPAs can be activated in pulses, making them yet more similar to biological muscle. However, braided pneumatic actuators could be limited not only by the complexity of control required to operate them, but also by hardware inherent to their system. Specifically, the valves which send compressed air to this biologically-similar muscle lose their ability to close at high spike frequencies, resulting in force saturation in the braided pneumatic actuators at high activation levels. We desired to determine whether these theoretical valve limitations pose an effect on the force output of the BPA. To do this, we generated pulse trains of varying inter-spike interval by injecting current into a spiking neuron simulated in SNS-Toolbox. These pulses were tested on two valve types to verify that a wide range of mean forces can be achieved in BPAs using spike activations. In addition, to test whether the activation curve could be modulated up to the saturation force in the muscle, a noise model was implemented in the spiking neuron that generated noisy spikes of activation in the BPA.
The human face is a powerful channel for emotional expression, making it central to how we communicate and connect. As robotics continues to evolve, a growing array of tools and applications are being developed to recognize and/or mimic facial expressions—bringing machines closer to truly intuitive, human-like interaction. However, the human face changes significantly over time due to aging. The impact of synthetic facial aging on emotion recognition has been underexplored. This study investigates how synthetic facial wrinkles—used to simulate aging on the synthetic faces—affect emotion perception in both human observers and automated facial expression recognition algorithms. We generated synthetic faces representing diverse ethnicities and genders, applying varying degrees of synthetic facial wrinkles. In Study 1, we validated the perceived age of these faces across ethnic and gender groups. In Study 2, we examined the recognition of six basic emotions (happy, surprise, fear, disgust, anger, and sad) by both human participants and a facial expression recognition algorithm. Our results show that while human participants remained relatively unaffected by aging in recognizing most emotions, the algorithm demonstrated significant performance variability, with increased wrinkle intensity improving the detection of certain emotions (e.g., disgust, anger) while degrading recognition of others (e.g., sad, surprise). These findings underscore the need for more diverse and age-inclusive training data in facial expression recognition systems and highlight the importance of considering facial aging effects in both emotion AI design and social robot development. We discuss the implications for creating more inclusive, emotionally intelligent robots that reflect the complexity of human aging.
Previous experiments using a biohybrid fly-robotic-interface (FRI) revealed that the neural activity of an identified directional-selective cell in the blowfly motion vision system indicates proximity to the walls of an experimental area. For a constant turning radius of the FRI along an oscillatory trajectory, the spike rate of the H1-cell was inversely proportional to wall distance. We have now implemented a distance estimation algorithm on a small 2-wheeled robot, based on a parsimonious model of directional motion detection in flies and tested its performance under different conditions. Applied to monocular and binocular image motion captured with an on-board video camera, the algorithm enabled the avoidance of collisions at a success rate of 93.5
Accelerating biohybrid research while ensuring safe and beneficial outcomes involves managing the tension between scientific freedom and social responsibility. Permissionless innovation, characterized by limited oversight and rapid experimentation, maximizes scientific freedom while downgrading consideration for future risks. In contrast, responsible innovation calls for ongoing stakeholder engagement, reflection, risk anticipation, social evaluation, and commitment to responsive design practices. While seemingly rooted in opposing ideologies – one reflecting risk tolerance, the other risk aversion – these methodologies share many similarities and can be integrated to afford scientific freedom during early-stage work while, as a project matures, providing strategies for ensuring and safeguarding societal benefit. By incorporating graduated oversight, the biohybrid research community has an opportunity to accelerate development and preserve safety, allowing investigators to capitalize on developing government interest.
Humans perceive the world through their bodies. The theory of object affordances suggests that when encountering an object, our brain encodes it not only based on its physical properties but also according to how we intend to use it. Decades of foundational research in neuroscience indicate that object properties are associated with distinct regions of the sensorimotor cortex, depending on the grasp type they tend to activate. In this study, we trained a Conditional Variational Autoencoder (CVAE) on the HO-3D_v3 dataset to reconstruct hand poses conditioned on object properties. Principal Component Analysis (PCA), clustering, and visualization of the model’s latent space revealed structured patterns for the abstract representation of the hand, which were distinctly organized according to object associations. This bears a notable resemblance to neural strategies observed in the human sensorimotor cortex for representing object-grasp relationships. This finding supports the notion that artificial intelligence systems can develop brain-like latent representations of object affordances. Such representations could significantly enhance robotic control in the future by enabling real-time motor planning for high-degree-of-freedom humanoid hand actions in an abstract latent space, bypassing the need for low-level pixel- and joint-level computations.
In the context of cybernetic systems and the future of human-AI collaboration, intelligent technologies are increasingly conceived as living machines: systems that regulate themselves through feedback while co-evolving with humans. Unlike traditional automation, these systems do not merely optimize fixed goals—they must continuously adapt in response to user behavior while preserving human agency, intent, and oversight. This paper presents the design and development of a Clinical Decision Support System (CDSS) for stroke neurorehabilitation that operationalizes these principles. While no experimental trials are reported in this manuscript, the system architecture and implementation are described in detail, and validation is planned through a multicenter randomized controlled trial (RCT). By embedding a feedback-driven, clinician-guided control loop into therapy delivery. Integrated within the Rehabilitation Gaming System (RGS), a virtual reality platform for adaptive motor and cognitive training, the CDSS maintains a dynamic model of each patient’s state, derived from behavioral performance, affective self-reports, and adherence patterns. This model drives the adaptive selection of therapeutic activities, forming structured, modifiable care plans. The system enables clinicians to inspect, adjust and refine interventions through a transparent interface, ensuring that algorithmic adaptation remains aligned with expert judgment and patient-specific needs. The CDSS exemplifies a shift from automation to co-regulation by supporting mutual adaptation between human and machine. It represents a concrete step toward the design of cybernetic healthcare systems, those capable of sustaining meaningful, personalized interactions over time, in service of complex and evolving recovery goals.
The data generated from sensors placed in and around complex engineering systems is increasing exponentially, and techniques are being developed to model the behavior of the system and learn about their failure and degradation over lifetime. Training these data driven models places requirements constraints on the data with regards to its availability, distribution and training process. The aim of the paper is to highlight the limitations of the model training methodologies and justify the need for a neuro-plausible framework for complex engineering systems. Viewing complex engineering systems as biological organisms facilitate in applying the learning from biological understandings to engineering problems. The paper explores engineering challenges from a biological standpoint, particularly through the lens of neuroscience, offering a biological perspective rather than a traditional engineering approach to solving complex engineering system problems.
Lobula Giant Movement Detectors (LGMD1 and LGMD2), neurons located in the locust’s optic lobe, are specialized in detecting approaching objects (looming perception) and have been widely modeled for integration into mobile robots. In bio-inspired robotic implementations of LGMD, inhibitory processes are crucial, as they help maintain selective responses to looming stimuli, enabling reliable collision avoidance. However, current robotic implementations of LGMD models often struggle with nearby translating movements, frequently generating false-positive collision alerts. Recent biological studies have identified trans-medulla afferent (TmA) neurons within the LGMD dendritic region, which may act as a form of self-inhibition (SI). These neurons rapidly suppress intermediate neuronal activities in situ within the LGMD structure, effectively complementing lateral inhibition (LI). Together, SI and LI enhance the specificity of looming responses, reducing interference from translating motions. Despite their biological significance, these mechanisms have yet to be effectively modeled and tested within artificial robotic vision systems. In response, this study introduces a biomimetic visual neural model that incorporates SI and coordinates it with LI during looming perception. The proposed neural computation explicitly activates SI during initial looming events and during translating movements by leveraging spatial correlations within segmented, localized image areas, defined as the local visual field (LVF). This innovative model has been integrated into a bio-inspired micro-robot, named Colias, serving as its sole collision detection mechanism. Both offline evaluations and real-world robotic tests demonstrate the efficacy of the biomimetic model in distinguishing looming from translating motions. Consequently, the robot exhibits significantly enhanced collision detection selectivity, closely resembling the capabilities observed in biological organisms.
This study investigates the dynamic behavior of braided pneumatic actuators (BPAs) under bio-inspired pulse modulation control to enhance their biomimetic performance. Building upon prior research on pulse timing and force amplification, we develop a state-space model to characterize the relationship between valve actuation and system pressure, optimized using Particle Swarm Optimization (PSO). A nonlinear model is then used to correlate pressure with force output via trust-region reflective least squares. Experiments were conducted on Festo BPAs at fixed lengths of 600 mm, 310 mm, and 140 mm, across various pulse frequencies and durations. Results show that the PSO-optimized model accurately predicts pressure behavior, with average errors of 5–16
Frogs use jumping as their primary mode of locomotion, but the force required for powerful jumps cannot be explained by leg muscles alone. In this study, we focused on Rana tagoi, a species capable of high vertical jumps, and investigated the interaction between leg kicking and trunk snap-through buckling. A musculoskeletal model was developed using MuJoCo, with muscle dynamics based on Hill-type models and anatomical parameters measured from actual specimens. Simulations showed that the model achieved high vertical jumps when both leg and trunk mechanisms were activated simultaneously, supporting previous results from frog cyborg experiments. These findings suggest that coordination between leg propulsion and trunk snap-through buckling is critical for vertical jumping. Future work will refine the model using detailed 3D bone geometry and quantify the role of elastic energy in jump performance.
Adaptations of motor dynamics represent a fundamental mechanism in verbal and non-verbal communication, enabling individuals to coordinate and adjust their behaviors during social interaction. These adaptations depend both on the characteristics of the entrainment signal and on external physical constraints, such as gravity and inertia. However, these external forces are often neglected in existing models of rhythmic coordination, which typically focus on temporal or neural aspects while overlooking biomechanical constraints. The objective of this study is therefore to analyze rhythmic adaptations within an environment that explicitly incorporates gravitational and inertial forces. The presented model is a force controller of muscle dynamics based on sensorimotor learning. Our first set of results demonstrates the capacity of the model to adapt its dynamics to desired velocities. Furthermore, by analyzing the energy expenditure of the system, we show that the model exhibits an asymmetric management of gravity between upward and downward movements which is consistent with observations reported in the literature. Altogether, these findings highlight the relevance and potential of our model for investigating biological movement generation, and in particular, movement rhythms and dynamics adaptations.
Bio-hybrid robotics—systems integrating living tissues with artificial mechanisms—challenge conventional ethical frameworks due to their ontological ambiguity and technological novelty. While some argue that ethics is either unnecessary or impossible in this domain—due to axiological pluralism, instrumentalist views of technology, or epistemic uncertainty—this paper rejects such deflationary positions. We argue that ethical governance in bio-hybrid robotics is both feasible and necessary, and that it can be grounded in a naturalistic theory of normativity informed by the evolution of cooperation in Homo sapiens. Drawing on game theory and the logic of collective action, we show that ethical failure in this domain is best understood as a problem of coordination under uncertainty: actors (researchers, institutions, and society) may endorse ethical principles privately, yet fail to act on them without common knowledge and mutual assurance. Using historical (chemical weapons, atomic research) and contemporary (CRISPR, generative AI) case studies, we demonstrate the consequences of ethical fragmentation and propose mechanisms for establishing shared ethical expectations, including public commitments, ethical observatories, and interoperable governance infrastructures. To avoid both ethical paralysis and ethical monoculture, we advocate for a model of pluralistic coordination grounded in evolutionary accounts of norm emergence and cognitive capacities for joint intentionality. Ethics, in this view, is not an external constraint but an infrastructural condition for responsible innovation. We use the term “post-biological” in the title not to imply the end of biology, but to signal a transition to systems in which biology is engineered, embedded, and functionally reconfigured in non-natural contexts.
