Traditional pedestrian detection methods based on red-green-blue (RGB) images struggle in adverse illumination, but a key capability required for pedestrian detection is all-day detection due to its critical role in diverse applications, e.g., security, surveillance, and autonomous driving. To address this issue, multispectral pedestrian detection attempts to introduce thermal images to supplement the RGB images, since they can be captured based on heat radiation difference without relying on external light sources. However, how to fuse the two modalities effectively is still lacking in-depth investigation. To prompt this field, we propose an implicit illumination-aware representation to address the limited availability of specific illumination labels in existing multispectral datasets, coupled with a prefusion feature alignment strategy to reconcile spatial misalignments of identical objects across modalities. We also identify four critical fusion challenges, revealing persistent limitations in existing multispectral detectors’ ability to holistically address these issues, particularly regarding underdeveloped cross-modal interactions and suboptimal cross-domain feature fusion. To this end, we propose a universal multispectral pedestrian detection paradigm (UMPDP), which includes a modality alignment module (MAM) for adaptive feature space alignment, a differential modality fusion module (DMFM) to enhance the relationship of different modalities, and a task-conditioned illumination module (TCIM) to dynamically adjust network weights based on illumination condition. Extensive experiments on KAIST and CVC-14 datasets demonstrate the general effectiveness of our proposed method. Code is available at https://github.com/gongyan1/UMPDP

Polymer-based composites attract increasing attention for gamma-ray shielding because of their light weight, flexibility, and easy processing. In this study, tungsten carbide-carbon (WCC) with dual-scale characteristics, composed of ultrafine WC1-x nanoparticles (3-4 nm) anchored on micrometer-scale carbon scaffolds, is incorporated into a polypropylene (PP) matrix to develop high-performance shielding materials. WCC possessing combined micro-and nano-scale features effectively mitigates nanoparticle agglomeration and enhances radiation attenuation across a wide energy range. Among the prepared composites, WCC45PP (45 wt % WCC) exhibits a linear attenuation coefficient (mu L) of 3.597 cm-1 and a radiation protection efficiency (RPE) of 48.3 % at 59.6 key, while also showing superior mechanical and thermal stability compared with the composite containing commercial WC. These results indicate that WCC with dual-scale characteristics provides an effective strategy for designing lightweight, flexible, and lead-free polymer composites for gamma-ray shielding applications.

Capturability characterizes a safe region of states for humanoid walking and is most commonly constructed by analyzing the one-dimensional divergent component of motion (DCM) of the center of mass. In this work, by exploiting the mathematical structure of the step-to-step (S2S) dynamics, we characterize capturability directly at the S2S level using the notion of a controlled positive invariant set (CPIS), which is computed via backward reachable sets (BRS). Although the inclusion of step time as a control input makes the S2S dynamics bilinear and renders these BRS difficult to compute directly, we show that, thanks to the specific mathematical structure of the S2S model, the capturable CPIS of the original nonlinear system can be exactly represented by the controlled invariant set of a linear system obtained at the minimum step time $T_{\min }$. Building on this set-valued characterization, we first propose a variable-step-time ALIP-based NMPC that employs the CPIS as a terminal set to guarantee recursive feasibility. We validated our approach through numerical simulations of the ALIP model, showing that an NMPC formulation equipped with a CPIS terminal set increases the feasibility (success) rate for challenging initial conditions near the boundary of the capturable set. We further demonstrated the effectiveness of the proposed method via both simulation and real-world hardware experiments on the Bruce humanoid robot, as well as simulation experiments on the Unitree G1.

Robots are playing an increasingly important role in tasks such as space assembly and manufacturing, but the extreme space environment makes them highly prone to failure. This work proposes a cooperative reconfiguration strategy for the modular robotic swarm, enabling the system to tolerate joint failures and sustain its manipulation capability autonomously. Notably, this process avoids the need for spare modules to replace faulty modules, which improves adaptability to the resource-constrained and extreme conditions of on-orbit missions. Especially under task space constraints, traditional fault-tolerant methods that rely solely on increasing redundancy remain ineffective, whereas cooperative reconfiguration can regenerate manipulation capability. A kinematic self-modeling method for modular robots is developed, leveraging topological representations and the product of exponentials (PoE) formula. After a failure occurs, the target configuration for reconfiguration is searched via particle swarm optimization (PSO), with the objective of maximizing the manipulation capability of the modular robot. A cooperative reconfiguration method considering joint constraints is developed on the basis of an improved rapidly exploring random tree (RRT) algorithm. This method enables the modular robot to transition from the initial faulty configuration to a new configuration capable of accomplishing the task. Experimental analysis of the assembly task validates that the proposed method can enhance the fault tolerance of a robot. In extreme cases involving multiple joint failures, the upper limit of fault tolerance for the number of faulty joints has been increased by 7 times. This fault-tolerant strategy holds significant potential for safeguarding the assembly and manufacturing capabilities of space robotic systems. A kinematic self-modeling method for variable-configuration robots is developed on the basis of screw theory and PoE.A fault-tolerant strategy for a modular robotic swarm to handle joint failures in extreme environments is proposed, using PSO and an improved RRT algorithm.Cooperative reconfiguration restores the manipulation capability of modular robots without spare modules.Cooperative reconfiguration remains effective under low-redundancy conditions and offers unique advantages under task-space constraints.

3D Gaussian Splatting (3DGS) has significantly advanced real-time novel view synthesis by representing scenes as dense collections of anisotropic 3D Gaussian primitives. However, the irregular spatial distribution of Gaussians often leads to poor GPU utilization, as warp divergence and redundant computation degrade rendering performance. To address this, we present Local-GS, a warp-coherent rendering paradigm that, organizes Gaussian primitives with respect to SIMT (Single Instruction, Multiple Threads) execution boundaries rather than scene geometry. Specifically, we propose three warp-coherent stages: a hoisting stage that precomputes shared parameters at tile level, a culling stage that discards warps with no contribution, and a blending stage that replaces per-pixel branching with a uniform instruction stream. Across extensive benchmarks on multiple datasets, Local-GS improves efficiency without compromising quality. As a plug-and-play optimization, it provides additional performance gains to all tested baselines, culminating in a $7.76\times$ speedup on Deep Blending scenes.

To mitigate critical tissue-injury risks arising from accumulated velocity-mode errors caused by controller sampling-interval uncertainties in teleoperated surgical robots with unmodeled dynamics, this paper proposes an adaptive control framework. Safe motion control is achieved through two key innovations: (1) Non-singular adaptive acceleration-gained control providing Cartesian acceleration compensation for direct end-effector force regulation, and (2) EKF-based state prediction to actively attenuate uncertainty effects. Rigorous simulations and experimental results validate the framework’s efficacy. The proposed method delivers robust transient performance (first peak overshoot < 0.2325 s) and high steady-state accuracy (reducing force error from 1.2741 mN to 0.5460 mN vs. nominal method), while effectively mitigating control input chattering during continuous motion. This approach thus enables high-precision velocity-mode control with enhanced stability and accuracy.

Precision motion control systems are frequently challenged by exogenous disturbances and plant uncertainties, which compromise tracking accuracy and operational stability. To address these issues, the extended state observer (ESO) has been developed to estimate both system states and disturbances in real time. However, conventional ESOs rely on an integral-based mechanism, necessitating a high observer bandwidth to handle rapidly varying disturbances. This, in turn, increases noise sensitivity and reduces stability margins. In this article, we introduce a novel error compensation mechanism that dynamically minimizes the estimation errors inherent in traditional ESOs, thus eliminating the need for a high observer bandwidth. Furthermore, we propose a time delay controller, leveraging historical input-output data to mitigate residual disturbances and plant variations. Experimental results validate the efficacy of our approach: settling time is reduced from 49.81 (conventional ADRC) to 2.74 ms (the proposed method), while disturbance convergence time decreases from 49.94 to 2.41 ms. These results underscore its superiority in high-precision applications that require robustness against disturbances and uncertainties.

Intracochlear theranostics, particularly targeted drug delivery and microsampling, offers a promising solution to inner ear diseases. However, specialized medical devices remain limited by a fundamental design challenge imposed by anatomical constraints: balancing miniaturization, dexterity, and perceptive functionality. Here, we present a low-aspect-ratio, dual-segment continuum robot that integrates catheter, endoscopic and instrumental functions. Driven by antagonistic cables, the robot uses a transition-free backbone composed of saddle-shaped joints to achieve a minimum bending radius of 1.9 mm. Dual-segment motion decoupling yields programmable C-/S-shaped configurations, facilitating anatomical navigation. The microneedle, embedded in the central channel, serves as an end-effector with positioning accuracy of 17.9 ± 4.1 μm. Fiber Bragg grating sensors, mounted on the needle, measure axial force to estimate tool-tissue interaction. Validation is performed on cadavers and in vivo animals, demonstrating the feasibility of a transcanal, atraumatic robotic paradigm. Thus, this system provides a practical and accessible approach for early diagnosis and treatment, helping extend precision medicine to underserved areas.

3D Gaussian Splatting (3DGS) enables real-time novel-view synthesis but remains limited on GPUs at high resolutions. Through a stage-wise Roofline characterization, we identify two distinct hardware bottlenecks: global memory traffic dominates the front end, whereas instruction throughput limits rasterization. Guided by this analysis, we develop RoofGS, a rendering framework that applies bottleneck-specific optimizations rather than generic kernel acceleration. For the memory-bound front end, we design a resolution-adaptive quantized depth sorting key that compresses each key to 32 bits. For the compute-bound rasterizer, we introduce a range-aware bit-level fast exponential approximation tailored to the bounded exponent range after opacity culling, with a derived per-pixel error bound. These two core techniques are complemented by additional optimizations (kernel fusion, compact attribute storage, culling, dual-pixel evaluation) that additionally reduce memory traffic and improve instruction-level parallelism. Experiments show that RoofGS achieves a 10.1× end-to-end speedup over 3DGS at 4K on an RTX 4090, increasing throughput from 61 to 616 FPS, with only a 0.028 dB PSNR loss.

This paper proposes an adaptive force-tracking controller for nonlinear systems with time-varying parameters, based on the practical task of peeling a film from the surface of a rigid body. The precise regulation of the peeling force by a force controller during robotic film peeling is critical for enhancing the safety and stability of the operation. Existing adaptive controllers for systems with time-varying parameters either result in large tracking errors or can only achieve closed-loop system stability, failing to track the desired force. The proposed controller employs the congelation of variables and robust adaptive control. The stability of the closed-loop system demonstrated via the corresponding Lyapunov function. Subsequent simulation results show that the tracking error of the closed-loop system with the proposed controller asymptotically converges to zero, validating the theoretical approach. Finally, experiments not only confirm the feasibility and stability of the proposed controller but also successfully demonstrate its potential for application in robotics.

For robotic assembly automation in intelligent manufacturing, high-precision peg-in-hole assembly constitutes a fundamental yet challenging task, where tiny pose errors can cause large contact forces and jamming. This paper proposes an ensemble robotic assembly skill-learning framework that integrates imitation learning (IL) and reinforcement learning (RL), with a geometric-constraint-based automatic pose refinement method to adapt a small number of human demonstrations to robot proprioceptive skill reproductions. By defining insertion-related variables, we train an offline multilayer perceptron (MLP) network to obtain the initial IL policy and then construct an ensemble RL architecture with an action layer generating soft actor-critic (SAC) exploratory actions and a parameter layer outputting hybrid force-position control coefficients and fusion weights that combine IL, hybrid force-position and SAC actions into task-execution commands. We first learn in simulation and then transfer to the real robot, where we first update the IL network with frozen RL parameters and subsequently fine-tune the RL policy with a few real-world episodes. Experiments demonstrate rapid sim-toreal convergence on round-hole assembly and robust cross-geometry transfer from round to triangular and square holes, as well as from square holes to USB insertion. Ablation studies and comparisons with stateof-the-art methods confirm improved sample efficiency, lower insertion forces, and higher success rates for cross-geometry generalization.

Linkage mechanisms with fewer closed loops exhibit limited enveloping angles, whereas multi-loop designs increase complexity, compromise reliability, and introduce structural interference issues. This paper establishes the kinematic general formula of the N-layer Reverse Four-Bar Linkage, whose spiral enveloping mechanism is inspired by the twining growth of climbing plants. It reveals the variation law of the envelope angle with the closed-loop layer number N, and explores the influence of structural parameters on the configuration. It is found that when the symmetric length conditions of the two sets of opposing links are satisfied and the three-pair links meet the internal-angle constraint α1=α2, the mechanism exhibits self-similar topological characteristics, allowing the mechanism to maintain kinematic stability during multi-layer expansion. In terms of prototype implementation, the multi-link interference issues were successfully addressed by adopting slotted shaft-thrust bearing composite joints and a stepped arrangement design, leading to the development of an N=6 six-layer Reverse Four-Bar Linkage prototype. The prototype achieves a theoretical envelope angle of 450°, enabling hyper form closure grasping. It can stably grasp objects such as cylindrical objects with diameters ranging from 35 mm to 110 mm, effectively adapting to the grasping requirements of targets with various sizes and shapes. This provides a highly versatile and reliable grasping solution for industrial automation scenarios.

Existing polar robots are constrained by limited energy supply, making it difficult to carry out long-term scientific exploration missions, which highlights an urgent demand for energy conservation. An energy-efficient multi-mode motion polar robot is proposed to address this challenge. Both increasing external assistance and reducing the driving force are critical for lowering energy consumption. A foldable sail is designed to provide external assistance. When unfolded, the sail generates assistive force. When folded, it maintains stability in extreme polar climates. The sail shape is designed based on a symmetrically extended NACA0018 airfoil, and the influence of different sail parameters on performance is discussed. The transformable tracks realize switching between traction and sliding modes through the separation of the track and teeth chain, using the sliding mode to reduce driving force. The effect of teeth parameter variations on traction performance is analyzed. The system kinematics and dynamics are model, and stability conditions are determined. Based on this, an energy-saving motion control framework for multi-mode motion is proposed. Finally, experiments are conducted to evaluate the energy-saving contribution of each independent mode under different configurations. Comprehensive experiments in multi-mode motion demonstrate an overall energy-saving rate of approximately 24%, verifying the effectiveness of the energy-saving motion control strategy. With its energy-saving advantages, this robot shows strong potential for enabling long-term scientific exploration in polar regions.

This paper proposes an intelligent compound disturbance rejection control framework integrating a novel Unknown System Dynamics Estimator (USDE) with Extreme Learning Machine (ELM). The USDE reconstructs the lumped term encompassing system parametric uncertainties and external disturbances online, requiring only real-time measurements of joint positions, velocities, and input torques, thereby eliminating dependency on a precise dynamic model. The framework further incorporates an ELM neural network to construct a disturbance rejection controller with direct joint torque actuation. Under randomly initialized ELM input weights, this architecture achieves effective prediction and compensation of acceleration errors through dynamic optimization of the output weights. Based on Lyapunov stability theory, the global stability of both the closed-loop tracking error and the USDE estimation error is rigorously proven. Simulations and experiments on a Franka Emika Panda robot demonstrate that the proposed method maintains high-precision trajectory tracking performance under simulated space disturbance scenarios, including unknown dynamic model mismatch, gravity variations, and sudden external disturbances. This work provides a theoretical framework and a universal implementation scheme, independent of precise dynamic models, for solving the challenge of fine manipulation control in harsh, unknown environments for open-space robotic systems.

The theoretical capability of modular robots to recover their original configuration or function after disintegration caused by external impacts has been cited as an advantage for planetary exploration. The key to achieving self-recovery lies in addressing the stochasticity of disintegration. Here, a self-recovery strategy is proposed for modular planetary exploration robots. Firstly, the recovery process is analyzed to construct a strategy framework and provide the problem definitions and strategy assumptions. Secondly, by standardizing the selection criteria for meta-modules under the stochasticity of disintegration, non-mobile modules can acquire mobility through the meta-module method, thereby laying the groundwork for executable self-recovery. Finally, a comprehensive optimization model is proposed, which encompasses the module interactions arising from stochastic disintegration. By integrating self-recovery characteristics with the simulated annealing algorithm, a solution method is designed to obtain self-recovery plans. Extensive hardware experiments were conducted, and the results demonstrate that the self-recovery strategy operates stably and executes successfully across various configurations and scenarios, thereby validating the feasibility and reliability. In this way, the self-recovery strategy and experiments could substantially advance the application of modular robots in space exploration, while also providing insights for other areas, such as assembly planning and applications involving non-mobile modular robots. Note to Practitioners-The motivation behind this paper is to address the autonomous recovery challenge of modular planetary exploration robots, and it also applies to other areas, such as efficient assembly planning and applications involving non-mobile modular robots. In extreme extraterrestrial environments, external impacts from collisions, meteorite impacts, and falls can cause robots to break apart into scattered fragments, resulting in task failure. By combining the separable and connectable characteristics of modular robots, we design a self-recovery strategy to efficiently and systematically reassemble scattered fragments. We propose meta-module selection criteria that endow non-mobile modules with mobility, thereby facilitating the execution of recovery plans. Additionally, we develop a mathematical model to generate recovery plans that coordinate the locomotion and interactions of various fragments during reassembly. Preliminary physical experiments suggest that the strategy is feasible, but current implementations rely on external localization devices. We are currently integrating inertial measurement units (IMUs) and ultra-wideband (UWB) ranging to achieve autonomous localization of the modules.

Trajectory planning plays a pivotal role in robotic motion planning, particularly in achieving time-optimal motion under complex dynamic constraints. Although the Time-Optimal Path Parameterization (TOPP) algorithm effectively addresses trajectory generation under joint torque constraints, classical methods often overlook third-order constraints. As a result, the generated trajectories, while torque-feasible, exhibit excessive jerk and poor dynamic stability, which limits their practical applicability. To overcome these limitations, this paper proposes a trajectory planning framework that simultaneously enforces torque and jerk constraints. Building upon torque-constrained TOPP, the method integrates a shooting-based strategy to identify switching points through bidirectional integration under jerk constraints and employs a Sigmoid-based fusion scheme to eliminate integration errors and ensure smooth transitions. The proposed approach is experimentally validated on a six-degree-of-freedom industrial robot. Comparative evaluations with the TOPP-RA algorithm demonstrate that the method significantly reduces both high-frequency vibrations during high-speed execution and residual oscillations after motion termination. Feedback from torque rate measurements, vibration sensors, and laser tracker data confirms faster settling and improved compliance, making the approach well-suited for complex industrial scenarios.

Cables are widely used in various fields such as automotive, electronics, and aircraft manufacturing. Cable following and insertion are fundamental tasks that are indispensable in numerous application scenarios. However, due to the high-dimensionality, flexibility, and deformability of cables, these tasks remain largely manual, suffering from high labor intensity, repetitive operations, and constrained workspaces, which motivates the need for automated solutions. This study proposes an automated framework for cable following and insertion based on tactile sensors, enabling robotic assembly of cables. During the cable following stage, a Transformer-based model is used to detect the cable’s sliding state to adjust the gripping force for smooth sliding within the gripper. Combined with a quadratic polynomial fitting to estimate the cable’s shape, this enables real-time adjustment of the cable pose inside the gripper, ensuring stable following while maintaining tension and preventing cable from sliding off. During the insertion stage, tactile sensing is used to monitor and regulate contact forces to guarantee successful insertion of the cable into a clamp. Experimental results validate the effectiveness of the proposed method and demonstrate its robustness across different cable specifications.

Quadrotor-assisted bipedal robots can improve locomotion stability by using redundant aerial thrust to enlarge the recoverable range of body attitude, center-of-mass motion, and external disturbances within which the robot can maintain or regain stable walking. However, unified modeling, control, and learning remain challenging due to closed-loop constraints, contact-switching discontinuities, and strong air-ground coupling. This paper proposes a unified framework integrating hybrid interpretable modeling, finite-time robust control, and evolutionary cooperative reinforcement learning. We first develop a multimodal composite linear inverted pendulum model, where an equivalent pivot unifies ankle-phase and toe-phase dynamics into a single representation, providing a convenient interface for planning and control. Based on this model, an orbital-energy gait planner is derived, and an adaptive-gain super-twisting nonsingular fast terminal sliding mode controller is designed, with gains scheduled by the error domain. This yields continuous hierarchical regulation: leg-level micro-adjustments for small deviations and rotor-assisted intervention for large deviations, with finite-time stability guarantees. Furthermore, we propose an evolution-enhanced cooperative proximal policy optimization (PPO) learning algorithm that combines parameter-space exploration via dynamic mutation adjustment with PPO refinement, augmented by an asymmetric Actor-Critic and a physics-consistency loss to improve stability and reproducibility. Simulations and prototype experiments demonstrate reduced attitude root mean square and long-tail destabilization risk, more concentrated energy consumption, fewer high-load events, and improved success rates, thereby also validating the reliability of quadrotor-assisted bipedal robot locomotion.

In dual-robot collaborative applications requiring high-precision trajectory synchronization, relative errors between robots often remain significant despite prior individual kinematic calibration using laser trackers. This paper identifies a critical issue wherein the robot pose transformation with respect to the tracker introduces projection discrepancies that are not reflected in the tracker-measured absolute errors. Consequently, although each robot exhibits sub-millimeter accuracy individually, contact-based measurements (via dial indicators) reveal relative errors of more than 2 mm when both robots execute identical circular Cartesian trajectories. It is discovered that the different servo performance cause asynchronous tracking, leading to movement direction-dependent relative deviation. To address this, we propose a novel game-theoretic iterative compensation strategy based on direct contact feedback. The relative errors are decomposed into motion corrections allocated to each robot according to a dynamic game model that reflects their error contribution and dynamic actuation capabilities. By applying iterative small-step trajectory adjustments, the system converges towards minimal relative error. Experimental results demonstrate the feasibility and superiority of the proposed method in achieving sub-millimeter relative accuracy. This study introduces a new paradigm that combines feedback-driven error decomposition with iterative learning and dynamic games, offering practical insights into high-precision dual-robot collaboration.

With growing real-world demands, efficient tracking has received increasing attention. However, most existing methods are limited to RGB inputs and struggle in multi-modal scenarios. Moreover, current multi-modal tracking approaches typically use complex designs, making them too heavy and slow for resource-constrained deployment. To tackle these limitations, we propose UETrack, a unified and efficient framework for single object tracking. UETrack demonstrates high practicality and versatility, efficiently handling multiple modalities including RGB, Depth, Thermal, Event, and Language, and addresses the gap in efficient multi-modal tracking. It introduces two key components: a Token-Pooling-based Mixture-of-Experts mechanism that enhances modeling capacity through feature aggregation and expert specialization, and a Target-aware Adaptive Distillation strategy that selectively performs distillation based on sample characteristics, reducing redundant supervision and improving performance. Extensive experiments on 12 benchmarks across 3 hardware platforms show that UETrack achieves a superior speed–accuracy trade-off compared to pervious methods. For instance, UETrack-B achieves 69.2% AUC on LaSOT and runs at 163/56/60 FPS on GPU/CPU/AGX, demonstrating strong practicality and versatility. Code will be made available.