The impact of eyes on attributions of agency and experience in humanoid robots.

Human action understanding serves as a foundational pillar in the field of intelligent motion perception.Skeletons serve as a modality- and device-agnostic representation for human modeling, and skeleton-based action understanding has potential applications in humanoid robot control and interaction. However, existing works often lack the scalability and generalization required to handle diverse action understanding tasks. There is no skeleton foundation model that can be adapted to a wide range of action understanding tasks. This paper presents a Unified Skeleton-based Dense Representation Learning (USDRL) framework, which serves as a foundational model for skeleton-based human action understanding. USDRL consists of a Transformer-based Dense Spatio-Temporal Encoder (DSTE), Multi-Grained Feature Decorrelation (MG-FD), and Multi-Perspective Consistency Training (MPCT). The DSTE module adopts two parallel streams to learn temporal dynamic and spatial structure features. The MG-FD module collaboratively performs feature decorrelation across temporal, spatial, and instance domains to reduce dimensional redundancy and enhance information extraction. The MPCT module employs both multi-view and multi-modal self-supervised consistency training. The former enhances the learning of high-level semantics and mitigates the impact of low-level discrepancies, while the latter effectively facilitates the learning of informative multimodal features. We perform extensive experiments on 25 benchmarks across across 9 skeleton-based action understanding tasks, covering coarse prediction, dense prediction, and transferred prediction. Our approach significantly outperforms the current state-of-the-art methods. We hope that this work would broaden the scope of research in skeleton-based action understanding and encourage more attention to dense prediction tasks.

Tactile sensors enabling human-like behavior to identify surface microstructures are essential for humanoid robots to interact precisely with complex environments. Most existing approaches use materials responding to dynamic forces and rely on machine learning methods to distinguish various types of surface microstructures. Quantitatively profiling the surface microstructures is significant but challenging, especially under the requirement of eliminating external bulky motion-control systems. Here, a quantitative tactile surface profiling strategy is presented through time-domain analysis of the signal of a piezoelectric twin-film architecture. The architecture uses two parallel piezoelectric films with a fixed interlayer distance, generating twin voltage signals with a time delay, which is inversely proportional to the scanning speed, and consequently removes the need for motion control. The microstructure heights correlate with the peak voltages, whereas widths and edge profiles are derived from the temporal analysis of distinct signal features. Tactile and in situ measurement of surface microstructures is demonstrated with high accuracy (>99.2%) over a broad height range of 1-1000µm. Furthermore, in-line quality inspection during additive manufacturing is realized by quantitatively profiling the surface microstructures. This work will drive innovations in tactile technologies that emulate and potentially surpass human capabilities and advance in situ surface characterization methods.

ABSTRACT As artificial intelligence technologies gain momentum in retail, humanoid service robots are becoming pivotal in transforming customer engagement. This study examines how the anthropomorphic features of these robots influence customer trust (cognitive and emotional), acceptance, and willingness to co-create value, with customer service expertise as a moderating factor. Drawing on the Computers as Social Actors theory and Service-Dominant Logic, the research proposes and tests a dual-pathway trust model, highlighting distinct mechanisms that mediate the relationship between robot anthropomorphism and retail outcomes. Data from a cross-sectional survey of 399 Chinese retail customers were analysed using SmartPLS 4.1. Findings reveal that robot anthropomorphism enhances both cognitive and emotional trust, which in turn drive acceptance and value co-creation. Crucially, customer service expertise asymmetrically moderates these pathways, strengthening the indirect effect through cognitive trust while leaving the emotional trust pathway unmoderated. Theoretically, the study enriches CASA and S-D Logic by disentangling the roles of dual trust mechanisms and identifying expertise as a critical, asymmetric boundary condition that governs the efficacy of cognitive-based social cues. In practice, the findings provide insights for the strategic designing and implementation of humanoid service robots to build trust, improve engagement, and facilitate culturally and expertise-sensitive value co-creation in the AI-enabled retail environments.

Humanoid robots are drawing significant attention as versatile platforms for complex motor control, human-robot interaction, and general-purpose physical intelligence. However, achieving efficient whole-body control (WBC) in humanoids remains a fundamental challenge due to sophisticated dynamics, underactuation, and diverse task requirements. While learning-based controllers have shown promise for complex tasks, their reliance on labor-intensive and costly retraining for new scenarios limits real-world applicability. To address these limitations, behavior(al) foundation models (BFMs) have emerged as a new paradigm that leverages large-scale pre-training to learn reusable primitive skills and broad behavioral priors, enabling zero-shot or rapid adaptation to a wide range of downstream tasks. In this paper, we present a comprehensive overview of BFMs for humanoid WBC, tracing their development across diverse pre-training pipelines. Furthermore, we discuss real-world applications, current limitations, urgent challenges, and future opportunities, positioning BFMs as a key approach toward scalable and general-purpose humanoid intelligence. Finally, we provide a curated and regularly updated collection of BFM papers and projects to facilitate further research, which is available at https://github.com/yuanmingqi/awesome-bfm-papers.

Background: Children with autism spectrum disorder (ASD) commonly experience persistent difficulties in social communication, emotional regulation, and social engagement. In recent years, artificial intelligence (AI)-based technologies, particularly socially assistive robots and intelligent sensing systems, have been explored as complementary tools to support psychosocial interventions in this population. Objective: This systematic review aimed to critically evaluate recent evidence on the effectiveness of AI-based interventions in improving social, emotional, and cognitive functioning in children with ASD. Methods: A systematic literature search was conducted in PubMed following PRISMA guidelines, targeting English-language studies published between 2020 and 2025. Eligible studies involved children with ASD and implemented AI-driven tools within therapeutic or educational settings. Eight studies met inclusion criteria and were analyzed using the PICO framework. Results: The reviewed interventions included humanoid and non-humanoid robots, gaze-tracking systems, and theory of mind-oriented applications. Across studies, AI-based interventions were associated with improvements in joint attention, social communication and reciprocity, emotion recognition and regulation, theory of mind, and task engagement. Outcomes were assessed using standardized behavioral measures, observational coding, parent or therapist reports, and physiological or sensor-based indices. However, the studies were characterized by small and heterogeneous samples, short intervention durations, and variability in outcome measures. Conclusions: Current evidence suggests that AI-based systems may serve as valuable adjuncts to conventional interventions for children with ASD, particularly for supporting structured social and emotional skill development. Nonetheless, methodological limitations and limited long-term data underscore the need for larger, multi-site trials with standardized protocols to better establish efficacy, generalizability, and ethical integration into clinical practice.

In a world where social interaction presents challenges for children with Autism Spectrum Disorder (ASD), robots are stepping in as allies in emotional learning. This study examined how affective interactions with a humanoid robot elicited physiological responses in children with ASD, using electrodermal activity (EDA) and heart rate variability (HRV) as key indicators of emotional arousal. The objectives were to identify emotionally salient moments during human-robot interaction, assess whether certain individual characteristics - such as age or ASD severity - modulate autonomic responses, and evaluate the usefulness of wearable devices for real-time monitoring. Thirteen children participated in structured sessions involving a range of social, cognitive, and motor tasks alongside the robot Pepper. The results showed that the hugging phase (HS2) often generated greater autonomic reactivity in children, especially among younger children and those with higher levels of restlessness or a higher level of ASD. Children with level 2 ASD displayed higher sympathetic activation compared to level 1 participants, who showed more HRV stability. Age also played a role, as younger children demonstrated lower autonomic regulation. These findings highlight the relevance of physiological monitoring in detecting emotional dysregulation and tailoring robot-assisted therapy. Future developments will explore adaptive systems capable of adjusting interventions in real time to better support each child's unique needs.

ABSTRACT Biological nociceptor is a key receptor to detect noxious stimulus and transmit warning signals to the central nervous system to trigger somatic responses. Artificial nociceptor can help to advance the development of humanoid robotics. A key feature of nociceptor is to sense pain on a graded scale and respond to it with varying degrees to avoid dangerous factors. However, artificial nociceptor with pain grading capability remains to be studied. In this work, an artificial nociceptor based on a memristor with multilevel quantized conductance states is reported. In addition to the critical “allodynia” and “hyperalgesia” features, the “pain grading” and “self‐healing” are demonstrated in our bio‐electronic nociceptive pathway by utilizing multilevel quantized conductance. Furthermore, nociceptive perception and avoidance response function is experimentally demonstrated with a bio‐electronic hybrid sensorimotor nerve by sending the synaptic feedbacks of the nociceptor to direct a sciatic nerve of mouse hind limb. Neuromodulation with increased muscle contraction and leg motion is achieved upon gradually increasing the pressure stimulation, mimicking the avoidance behavior. This work proposes a promising artificial nociceptor for developing humanoid robotics.

Humanoid robots often employ rigid-trunk designs which limit locomotion capabilities and payload capacity. This paper presents a novel bio-inspired, tensegrity-based flexible spine mechanism designed to address these limitations. The design integrates a modular, multi-segment structure combining rigid struts and flexible TPU cables, creating a compliant, stable, and adaptable spine. We developed a novel dynamic model of this tensegrity-based spine to analyze its motion characteristics, providing detailed insights into its load-bearing capabilities and range of motion. Experimental results, obtained using a novel humanoid robot platform ("Flexinoid"), demonstrate improvements in locomotion performance. Furthermore, the design mitigates non-linear movement challenges, allowing for an enhanced range of flexion to -30°:65° and lateral bending by ± 30°. The experimental results confirm an increase in sensitivity and a decrease in the minimum detectable payload following the onset of motor back-driving, validating the effectiveness of the passive energy storage mechanism during initial loading. This enhancement in performance underscores the potential of this bio-inspired design for applications requiring precise control and high payload capacity. This research presents a novel approach to humanoid robot design, paving the way for more versatile and capable robots in various applications.

For a humanoid robot, it is difficult to predict a motion trajectory through end-to-end imitation learning when performing complex operations and multi-step processes, leading to jittering in the robot arm. To alleviate this problem and reduce the computational complexity of the self-attention module in Vision-Language-Action (VLA) operations, we proposed a memory-gated filtering attention model that improved the multi-head self-attention mechanism. Then, we designed a cross-modal alignment perception during training, combined with a few-shot data-collection strategy for key steps. The experimental results showed that the proposed scheme significantly improved the task success rate and alleviated the robot arm jitter problem, while reducing video memory usage by 72% and improving training speed from 1.35 s to 0.129 s per batch. This maintained higher action accuracy and robustness in the humanoid robot.

Hybrid constrained motion planning scheme for wheeled mobile humanoid dual-arms robot

This study presents a low-cost 15-DOF humanoid robot designed using Arduino to overcome the limitations of existing humanoid systems that rely on complex controllers, high-end processors, and computationally heavy whole-body coordination methods. The objective is to develop a compact, efficient humanoid capable of stable walking, turning, and head-tracking using lightweight motion planning and sensor-based corrections. The proposed method integrates simplified keyframe gait generation, servo-angle computation, and a real-time feedback loop implemented through Arduino IDE, enabling precise actuation across 15 metal-gear servos. Experimental evaluation shows improved servo angle accuracy by 12% and enhanced motion stability by 9% compared to baseline manual-tuned control. No external dataset is required, as all data are generated from robot trials. Results confirm that the proposed architecture ensures reliable, repeatable movement, offering a promising foundation for compact humanoid development.

Social robots are increasingly integrated into early childhood education, yet limited research exists examining preschoolers’ learning from robotic versus human demonstrators across embodied tasks. This study investigated whether children (aged between 3 and 6) demonstrate comparable performance when learning body-centered tasks from a humanoid robot compared to a human demonstrator. Sixty-two typically developing children were randomly assigned to a robot or a human condition. Participants completed three tasks: body part comprehension and production, body movement imitation, and emotion recognition from body postures. Performance was measured using standardized protocols. No significant main effects of demonstrator type emerged across most tasks. However, age significantly predicted performance across all measures, with systematic improvements between 3 and 6. A significant age × demonstrator interaction was observed for sequential motor imitation, with stronger age effects for the human demonstrator condition. Preschool children demonstrate comparable performance when interacting with a humanoid robot versus a human in body-centered tasks, though motor imitation shows differential developmental trajectories. These findings suggest appropriately designed social robots may serve as supplementary pedagogical tools for embodied learning in early childhood education under specific conditions. The primacy of developmental effects highlights the importance of age-appropriate design in both traditional and technology-enhanced educational contexts.

Humanoid robots with advanced sensory capabilities are increasingly demanded for empathetic, close-contact interactions with humans. Electronic skin (E-skin) is a key enabling technology for such tactile perception. However, current E-skins are limited to basic tactile sensing with simple circuit architecture. Here, we introduce a neuromorphic robotic e-skin (NRE-skin) that not only provides fundamental tactile sensing but also integrates advanced features, such as active pain and injury detection. The NRE-skin encodes dynamic tactile stimuli into neural-like pulse trains and features active pain detection that triggers protective reflexes. Additionally, its injury sensing and modular design enable precise localization of damaged areas and rapid replacement of affected skin. By emulating human sensory and protective systems, the NRE-skin facilitates more natural and safer human-robot interactions.

The design and implementation of a suitable robotic hand for a toddler-sized humanoid robot is a challenging task. The main purpose of this work is to optimize the design of an anthropomorphic robotic hand and control it by using surface electromyographic (sEMG) signals. Isolation forest backward particle swarm optimization is used to optimize the robotic hand. The fitness function is defined by thumb opposability and the ability to grasp objects based on grasp taxonomy. Learning without forgetting (LWF) is adopted to train sEMG signal data sequentially, and the consequently learned model is used as an ensemble to control the optimized robotic hand. Webots is adopted to simulate the scenario of grasping objects to optimize the design of the hand. The optimized robotic hand is compared with two robotic hands, and the highest fitness values in the simulator and real world are obtained. Three different sEMG inputs, namely, raw data, bandpass, and discrete wavelet transformed bandpass, are compared in LWF, and the structure of neural networks is considered. The final LWF model is successfully applied to a real-world system to manipulate a robotic hand via hand gesture classification in real time.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-30658-2.

Humanoid robots operating in unstructured environments and under high-load conditions commonly face challenges such as limited locomotion performance and the difficulty of balancing structural strength with weight reduction. This study proposes a novel bio-inspired electro-hydraulic humanoid robot that incorporates a parametric dynamic model based on the coupled muscle-tendon-bone characteristics of the human hip-knee-ankle complex. Leveraging a custom-designed, reverse-inverse kinematics framework, the leg morphology and electro-hydraulic actuator parameters are co-optimized to enhance agility and obstacle-crossing capabilities. To simultaneously ensure structural strength and mass control, honeycomb structures are designed for the leg components, achieving functional lightweighting while preserving balanced strength across different directions. Simulation analyses demonstrate that a 21.28% weight reduction is attainable while maintaining comparable out-of-plane equivalent elastic and shear moduli relative to the original structure, thus meeting the demands of complex loading and impact conditions. Experimental tests confirm that the robot exhibits robust environmental adaptability and stable locomotion during high-speed running at 10km/h and obstacle traversal over 300mm. The findings validate the effectiveness of the proposed configuration and bio-inspired strategy, providing theoretical support and an engineering paradigm for structural optimization and system integration in high-performance humanoid robots under complex task scenarios.

The rapid advancement of humanoid robotics has spurred researchers’ interest in flexible sensors for wide linear range detection. In response, we report a capacitive flexible pressure sensor based on a multi-walled carbon nanotubes/titanium dioxide/polydimethylsiloxane (MWCNTs/TiO2/PDMS) composite. A micro-hemispherical structure array formed on the composite surface via a templating method reduces the initial capacitance value. Modified carbon nanotubes (F-MWCNTs) were prepared using 2 wt%, 5 wt% and 10 wt% γ-aminopropyltriethoxysilane (APTES), significantly enhancing dispersion and interfacial bonding strength. The synergistic effect of microstructures and MWCNTs surface functionalization further enhances sensing performance. The F-MWCNTs/TiO2/PDMS pressure sensor modified with 2 wt% APTES exhibits outstanding sensing capabilities: it demonstrates dual-stage sensitivity across a broad linear range of 0–95 kPa (0–13 kPa segment: 1.89 ± 0.49 kPa−1; 13–95 kPa segment: 7.08 ± 0.63 kPa−1), with a response time of 200 milliseconds, maintaining stability over 2500 cyclic loadings. In practical application exploration, this sensor has demonstrated strong adaptability, confirming its significant potential in micro-pressure detection, wearable electronics, and array sensing applications.

Abstract This article asks a deep and urgent question for the age of artificial intelligence: how will the gains from algorithmically mediated production be distributed across society? We reconceptualize algorithmic surplus value (ASV) not as a break with Marx’s labour theory of value but as a digital-era intensification of relative surplus value. On our account, AI systems—still constant capital or “dead labour”—reorganise production and circulation by compressing socially necessary labour time and by enclosing informational and infrastructural rents; they do not autonomously create value. Building on this theoretical repositioning, we bridge principles and practice by proposing operational tools—value-based filters and culturally responsive value repositories—that parameterize fairness, accountability, and pluralism within algorithmic pipelines. We clarify the role of large language models as assistants, not oracles, suitable for analysis and scenario generation but not for moral adjudication. We then outline institutional pathways—public data funds, transnational minimum-threshold and options-menu compacts, and policy sandboxes with auditable metrics—through which ASV’s gains can be steered toward public ends. Frontier modalities in humanoid care robotics, quantum machine learning, and neuromorphic and edge computing are used as stress tests to show how coordination speedups and rent enclosure can widen distributive asymmetries absent governance. The contribution is both analytical and practical: we couple a Marxian account of value extraction to implementable mechanisms and institutions so that, as AI scales, algorithmic wealth becomes a common good rather than a private windfall.

Abstract In this work, a robust control design for balance and walking control of humanoid biped robot in presence of external disturbances is proposed. The controller combines Capture Point (CP) tracking controller with Extended State Observer (ESO) in order to achieve robustness against unknown disturbances, parametric uncertainties as well as unmodeled dynamics denoted as a composite disturbance. The effect of the composite disturbance is estimated by ESO as an extended state of the system and the estimate is used to augment the CP tracking controller in order to achieve robustness. Closed loop stability analysis under the proposed control is carried out. The efficacy of the design is first verified for a simplified Linear Inverted Pendulum Model (LIPM) and subsequently, through NAO multi-body simulations. It is shown that the proposed design achieves robust balance and walking control for biped humanoid robot in presence of unknown composite disturbances.

How humanoid robots influence consumer preferences in the foodservice industry.