The seamless integration of mobile robots into daily life hinges on addressing the challenge of human–robot interaction within dynamic environments. Existing research has extensively explored the harmonious path planning of robots with individual pedestrians. However, the practical scenario often involves robots navigating among pedestrian groups. In such contexts, due to inadequate consideration of collective pedestrian dynamics, conventional approaches frequently lead to group disruptions, structural instability, and rigid interactions. This study introduces a novel human–robot collaborative navigation framework on the basis of the interactive information entropy, which is proposed to quantify information about pedestrian groups, assesses the capacity of robots to acquire group-related information from different positions, and offers a robust decision-making basis for identifying optimal interaction points. The methodology comprises three key phases: first, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm is employed to identify pedestrian groups by analyzing real-time positional coordinates and kinematic parameters; second, scenarios are categorized into dynamic and static types based on pedestrian states, with candidate interaction points generated according to the distribution and orientation of pedestrians; finally, an interactive information entropy model is designed to evaluate and select the optimal interaction point, guiding trajectory planning of the robot while respecting group coherence and social navigation conventions. Extensive evaluations in both simulated and real-world environments verify the feasibility and effectiveness of this approach.

Efficient and accurate underwater target recognition is crucial for assessing the stability of marine ecosystems. However, traditional manual diving surveys are costly, time-consuming and limited in coverage. With the advent of new AI technologies, intelligent underwater robots have emerged as a novel solution to this problem, but their perceptual capabilities usually rely on extensive and accurate annotated data, which is extremely challenging for underwater datasets with poor imaging quality and scarce pixel-level annotated samples. To address the above issues, this paper proposes a semi-supervised semantic segmentation framework specifically used for underwater ecological monitoring. Specifically, a multi-modal perturbation consistency method based on a multi-branch structure is constructed to simultaneously process multiple input modalities and extract intermediate features. A novel pseudo-label screening strategy based on semantic similarity is proposed, which utilizes the mapping relationship between labeled and unlabeled images to screen reliable pseudo-labels. Experiments were conducted on three underwater scene datasets, and tests were carried out on our self-developed underwater platform. The mIoU values on the SUIM dataset under 1/2 and 1/4 splits reached 72.06% and 71.45% respectively, outperforming the existing mainstream methods. In the future, this method can be deployed on underwater platforms to provide technical support for the implementation of automated and large-scale marine ecological monitoring systems. The codes are available at: https://github.com/A0268/uwea.

Parallel robotic gripper is an efficient tool for grasping and manipulating objects in industrial applications. In recent years, to enable adaptive grasping of objects with complex shapes, many parallel grippers are equipped with soft robotic fingers. However, due to the low structural stiffness of the utilized soft materials, the grasping payload of the soft fingers is usually limited. To cope with this issue, we propose a topology-optimization-based method in this article to enhance the grasping payload of soft fingers for parallel grippers. Using a multiobjective algorithm, the adaptive grasping ability of the monolithic finger and the holding stiffness of the fingertip are taken into account in the optimization procedure. The realized finger is fabricated with thermoplastic polyurethane (TPU) material. To evaluate the soft-rigid hybrid performance of the synthesized finger, stiffness tests and grasping payload tests are also conducted. Experimental results show that the optimized finger has a higher payload capacity than the conventional finray-like soft finger, while maintaining similar adaptive grasping properties. Furthermore, a series of grasping tests have also demonstrated the grasping adaptability of the synthesized soft robotic finger for objects with different materials, shapes and weights.

Whole-body reactive obstacle avoidance for mobile manipulators (MM) remains an open research problem. Control Barrier Functions (CBF), combined with Quadratic Programming (QP), have become a popular approach for reactive control with safety guarantees. However, traditional CBF methods often face issues such as pseudo-equilibrium problems (PEP) and are ineffective in handling dynamic obstacles. To overcome these challenges, we introduce the Adaptive Cyclic Inequality (ACI) method. ACI takes into account both the obstacle’s velocity and the robot’s nominal control to define a directional safety constraint. When added to the CBF-QP, ACI helps avoid PEP and enables reliable collision avoidance in dynamic environments. We validate our approach on a mobile manipulator comprising both low-dimensional and high-dimensional subsystems, demonstrating the generality of the framework. In addition, we integrate a simple yet effective method for avoiding self-collisions, allowing the robot enabling comprehensive whole-body collision-free operation. Extensive benchmark comparisons and experiments demonstrate that our method performs well in unknown and dynamic scenarios, including difficult tasks like avoiding sticks swung by humans and rapidly thrown objects.

Semantic segmentation is a crucial technology for intelligent vehicles, enabling robust scene understanding in complex driving environments. However, existing methods often struggle with small, distant, and overlapping objects, posing challenges for safe autonomous operation. To address these limitations, we present FineSegNeRF, a model designed for fine semantic segmentation of such challenging scenarios. Our approach separates features along the depth dimension to perceive stereoscopic scene from spatial dimension, and then uses NeRF’s multi-view consistency to optimize the separated features for fine understanding. Meanwhile, a new “Semantic Uncertainty Neural Volume Render” method is proposed for constraining the consistency of volume density and semantic uncertainty estimation to further improve the semantic segmentation performance. Compared to current representative RGB-D and NeRF fusion semantic segmentation methods, our approach performs remarkable competitiveness in terms of fine semantic segmentation on VKITTI 2 and Replica datasets.

With the rapid advancement of tactile sensing technologies, real-time prediction of robotic grasp stability has become increasingly critical for ensuring safe and reliable operation in automated industrial environments. Achieving accurate real-time predictions of grasp instability is essential for preventing failures during robotic manipulation tasks. Tactile signals serves as a primary source of contact information during grasping and provides detailed insight into the interaction forces and surface dynamics. However, tactile signals often exhibit complex and non-stationary temporal characteristics, posing significant challenges for efficient and robust modeling. To address these challenges while maintaining computational efficiency, we propose GraspLite-Net, a lightweight deep learning-based model that relies solely on tactile time-series data, without requiring visual input. Our proposed GraspLite-Net captures both local and global temporal dependencies while remaining suitable for real-time deployment and comprises three main components: (1) a deformable patch segmentation module that adaptively identifies informative temporal segments; (2) a multi-scale re-parameterized convolutional block for enhanced temporal feature extraction; and (3) an attention-based multiple-instance learning (MIL) pooling module for effective feature aggregation. We conduct extensive experiments on the BiGS dataset and show that GraspLite-Net surpasses baseline methods in both prediction accuracy and inference speed, underscoring its effectiveness and suitability for deployment in real-world robotic grasping systems. We further perform a cross-dataset evaluation on the SnapFitForceProfiles dataset to demonstrate the model’s generalization capability to new robotic scenarios.

Data-fusion networks have shown significant promise for RGB-thermal scene parsing. However, the majority of existing studies have relied on symmetric duplex encoders for heterogeneous feature extraction and fusion, paying inadequate attention to the inherent differences between RGB and thermal modalities. Recent progress in vision foundation models (VFMs), which leverage self-supervised learning on large-scale unlabeled datasets, has exhibited superior capabilities in extracting informative, general-purpose features compared to supervised encoders. However, their potential has yet to be fully leveraged in the domain. In this study, we take one step toward this new research area by exploring a feasible strategy to fully exploit VFM features for RGB-thermal scene parsing. Specifically, we delve deeper into the unique characteristics of RGB and thermal modalities, thereby designing a hybrid, asymmetric encoder that incorporates both a VFM and a cross-modal spatial prior descriptor (CSPD), enabling enhanced extraction of complementary heterogeneous features. The extracted features undergo dual-path feature fusion through our proposed progressive heterogeneous feature integrators. Moreover, we introduce an auxiliary task to further enrich the local semantics of fused features, thereby improving the overall performance of RGB-thermal scene parsing. Our proposed HAPNet, incorporating all these components, delivers superior performance under challenging illumination conditions. Extensive experiments demonstrate that HAPNet outperforms all other state-of-the-art methods, with improvements of 0.1%, 1.0%, and 2.4% in mIoU on three public RGB-thermal scene parsing datasets: MFNet, PST900, and KP Day-Night, respectively. Additionally, our method exhibits exceptional generalizability for RGB-HHA scene parsing. We believe this new paradigm has opened up new opportunities for future developments in data-fusion scene parsing approaches. The source code is publicly available at https://mias.group/HAPNet/.

Flying fish, as a quintessential cross-medium species, are well known for their exceptional underwater swimming and water–air transition capabilities. However, their aerial gliding kinematics — particularly the influence of pectoral fin morphology on gliding performance — remains insufficiently explored. In this study, we develop a bio-inspired gliding model to investigate airborne gliding characteristics. Four pectoral fin designs based on distinct biological morphologies (Model A: flying fish; Model B: devil ray; Model C: puffin; Model D: pigeon) were fabricated, and their aerodynamic performance was evaluated through hand-throw tests, catapult-based gliding experiments, and CFD simulations. By varying the angle of attack and launch velocity, we systematically analyzed their effects on gliding efficiency, while numerical simulations were conducted to further assess aerodynamic forces and predict gliding dynamics. Results show that under a 20° angle of attack and a launch speed of 13 m/s, the flying fish inspired fin (Model A) achieved a maximum gliding distance of 5.97 m, which is 8.74%, 10.97%, and 11.38% longer than the other three models. The launch energy required per unit gliding distance was also reduced by 8.83%, 12.36%, and 11.52%, respectively. Both experimental and numerical findings confirm the superior aerodynamic performance of the flying fish pectoral fin in aerial gliding. This study provides insights for the future development of bio-inspired flying fish robots capable of air gliding.

A central challenge in bio-robotics is to create machines that can integrate into and illuminate natural ecosystems. The Chinese Yangtze Alligator – a critically endangered species exhibiting exceptionally agile spine-leg coordination honed by its terrestrial-aquatic transition – offers​ a unique model to address this challenge. Yet, existing alligators-like robots fail to capture such biological fidelity due to insufficient actuation, simplified mechanics, and the absence of adaptive control policies. Here, we introduce the Spine-Legged Adversarial Imitation and Reinforcement Learning (SLAIR) framework, which for the first time leverages deep reinforcement learning to master this coordination. By retargeting biological motion data from Yangtze alligators and integrating impedance control to produce natural compliance, our controller achieves adaptive spine-leg coordination in a custom 24-DOFs robot. A variational autoencoder (VAE) generalizes across terrain, while a dual-critic architecture robustly fuses imitation and task rewards. This enables agile locomotion (0.32 m/s, 360° turns in 3.5 s) with a 46.7% reduction in cost of transport. Crucially, the robot’s biomimetic fidelity was validated in the field, where it elicited natural curiosity and approach behavior from wild Yangtze alligators—demonstrating its potential as a transformative tool for conservation biology.

Flapping-wing micro aerial vehicles (FWMAVs) offer compelling advantages for navigating confined and cluttered environments due to their low weight, agility, and bioinspired locomotion. This study introduces a novel quad-wing FWMAV architecture integrating a spatial dual-crank rocker mechanism that enables active thrust-vector-based yaw steering via independently driven rear wings, eliminating reliance on drag-inducing surfaces. The platform also incorporates a lightweight digital shadow system for real-time telemetry, state visualization, and autonomous feedback control without dependence on external tracking infrastructure. Three variants were fabricated and tested: a baseline model (QF-Base), an endurance-optimized model (QF-L), and a digital shadow-enabled model (QF-DS). Experimental evaluations demonstrate superior yaw maneuverability, altitude retention during turns, and extended endurance — up to 16 min 13 s with payload — while digital shadow integration enables autonomous monitoring in GPS-denied scenarios. The results establish a new framework for energy-efficient, sensor-integrated FWMAVs, addressing key limitations in current platforms and advancing the capabilities of next-generation bioinspired aerial robotics.

As super minimally invasive surgery (SMIS) evolves in clinical practice, interdisciplinary integration has emerged as the driving force behind its advancement. The natural orifice flexible endoscopic surgical robot represents a technological platform integrating advanced technologies, including mechanical metamaterials, multi-degree-of-freedom design, and intelligent control systems. This integration demonstrates the benefits of interdisciplinary collaboration, enabling precise, traceless SMIS. Designed specifically for narrow and curvilinear organ lumens, the robot’s microstructure-driven design, highly dexterous mechanical structures, and intelligent control systems enhance the precision, flexibility, and stability of surgical procedures, surpassing limitations of traditional endoscopic techniques in managing complex lesions. This article summarizes key technological advancements, highlighting how microstructure design, dexterous mechanical structures, and intelligent control systems improve surgical safety, accuracy, and effectiveness. We further analyze the critical role of artificial intelligence in system modeling, control strategies designing, and multimodal sensory information processing. Additionally, key non-technical aspects are discussed, such as standardized surgical training processes, system safety, regulatory frameworks, and clinical promotion strategies. In summary, this article aims to provide a comprehensive framework for interdisciplinary research involving materials science, mechanics, artificial intelligence, clinical medicine, facilitating their progression from prototypes to widespread clinical applications in SMIS robots.

The demand for large-scale space infrastructures, such as solar power stations and ultra-large telescopes, is growing rapidly. Consequently, on-orbit assembly has emerged as a critical pathway to overcome launch vehicle constraints and construct next-generation space assets. The connection interface, serving as the core of any on-orbit assembly system, directly determines the performance of the assembled structure, mission reliability, and system scalability. This paper provides a systematic review of the technological developments in on-orbit assembly interfaces. We propose an innovative four-category classification framework: Rigid Structural Connections, Mechanism-Based Connections, Advanced Integrated Interfaces, and Permanent In-Orbit Fixation. This framework clearly delineates the technological spectrum from high-stiffness mechanical joints to material-level integration. By analyzing the design principles, typical examples, and technical characteristics of each category, this paper reveals the trade-offs in task adaptability among different interfaces. It thereby offers a theoretical foundation and technical reference for the future development of standardized, autonomous, and multifunctional interfaces within the In-orbit Servicing, Assembly, and Manufacturing (ISAM) ecosystem.