Journal of Bionic Engineering最新文献

{"title":"Nonlinear Dynamic Analysis of a Novel Multi-coupled Quadrupedal Bio-inspired Vibration Isolation Platform","authors":"Shihua Zhou,&nbsp;Xinpeng Xu,&nbsp;Zeyao Mu,&nbsp;Pengyang Wang,&nbsp;Ye Zhang,&nbsp;Yulin Liu,&nbsp;Zhaohui Ren","doi":"10.1007/s42235-026-00973-z","DOIUrl":"10.1007/s42235-026-00973-z","url":null,"abstract":"<div><p>In order to achieve ultra-low and broadband frequency vibration suppression, a structurally compact and parameter-adaptable Multi-coupled Quadrupedal Bio-inspired Vibration Isolation platform (MQBVI) based on the mechanical features of the quadruped is proposed in this research. Due to the large working stroke and strong nonlinearity of the platform, the virtual work method and incremental harmonic balance method are utilized to systematically analyze the static and dynamic characteristics, respectively. The comparison validations of bio-inspired platform are conducted by Incremental Harmonic Balance Method (IHBM), ADAMS, numerical method and physical prototype. The investigation revealed that the restoring force and equivalent stiffness are nonlinear and asymmetrical, which lead to the MQBVI platform exhibits superior High Static Low Dynamic Stiffness (HSLDS) characteristics and strong stability. The vibration suppression performance of the MQBVI structure is investigated with installation angle, spring stiffness, connecting rod length, isolated mass and excitation amplitude, which exhibits low resonance frequency and wide effective vibration attenuation frequency band. Notably, the minimum critical values of resonance frequency and vibration suppression effect are reduced to 0.27 Hz and 0.39 Hz with <i>α</i> = 30°, respectively. Overall, the proposed MQBVI platform is an efficient ultra-low frequency vibration suppression structure with excellent adaptability, and the bio-inspired vibration buffering and isolation mechanism are revealed.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"2918 - 2934"},"PeriodicalIF":4.4,"publicationDate":"2026-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903609","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Research on the Selection and Rapid Positioning of Facial Drive Points for Humanoid Robot Heads by Characterizing Human Facial Expressions","authors":"Zhibin Yan,&nbsp;Jiayu Zhao,&nbsp;Zhouyi Wang,&nbsp;Liuwei Wang,&nbsp;Zhendong Dai","doi":"10.1007/s42235-026-00969-9","DOIUrl":"10.1007/s42235-026-00969-9","url":null,"abstract":"<div><p>Achieving realistic facial expressions is a central goal in humanoid robot head research. Current methods for determining facial drive points, often based on FACS and design experience, are subjective and prone to bias. This study proposes a quantitative method for selecting facial drive point positions based on optical motion capture and interpolation reconstruction, providing a set of 3D coordinates. On this basis, we further propose a rapid positioning method for robotic drive points applicable to diverse human bionic subjects. Motion data of 238 reflective markers were captured from volunteers performing six basic and one arbitrary facial expression to represent facial movement characteristics. Drive areas were defined based on facial muscle distribution and movement patterns, and the maximum displacement point in each area was calculated using TPS interpolation and displacement field reconstruction to obtain full-face drive point coordinates. For diverse human bionic subjects, proportional correspondences between their inherent facial features and the drive point parameters established in this study were calculated, and new facial drive points were positioned through scaling. To validate the approach, based on the results, a 19-degree-of-freedom humanoid head prototype was built, realizing six basic expressions with recognition rates exceeding 90%, thereby validating the feasibility of the proposed drive point determination method for practical applications. In summary, the proposed methods provide a quantitative reference for subsequent research, enhance design efficiency, and provides a theoretical foundation for achieving more natural and realistic robotic facial expressions.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"3001 - 3019"},"PeriodicalIF":4.4,"publicationDate":"2026-08-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903535","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"QLOA: A Self-Adaptive Q-Learning-Based Optimization Algorithm with Dynamic Mathematical Formulation Selection","authors":"Najibeh Farzi-Veijouyeh,&nbsp;Mohammad-Reza Feizi-Derakhshi","doi":"10.1007/s42235-026-00961-3","DOIUrl":"10.1007/s42235-026-00961-3","url":null,"abstract":"<div><p>In the field of optimization, one of the fundamental challenges of existing methods lies in their inability to dynamically adapt to the diverse characteristics of objective functions. Many algorithms rely on a fixed optimization strategy; therefore, while they may perform well on certain problems, they often lose efficiency when dealing with functions exhibiting different behaviors, such as smooth or steep landscapes. To overcome this limitation, this study introduces a smart and self-adaptive approach called the Q-Learning-Based Optimization Algorithm (QLOA). The proposed QLOA incorporates a set of diverse mathematical formulations designed to enhance both the exploration and exploitation phases. A key feature of QLOA lies in the integration of the Q-learning framework, which enables the algorithm to intelligently and adaptively select the most suitable formulation based on the problem characteristics and its performance feedback, thereby establishing a dynamic balance between exploration and exploitation at each stage of the optimization process. In addition, population refinement mechanisms—such as eliminating redundant members and applying structural adjustments—prevent the algorithm from becoming trapped in local optima. To further enhance adaptability in problems with hard and soft constraints, an extended version named QLOA with Constraint Management (QLOA-CM) is developed. Experimental results on standard benchmark functions and the CEC2019 and CEC2022 test suites demonstrate that QLOA outperforms reference algorithms in both the Friedman and Wilcoxon statistical tests, confirming its efficiency, robustness, and statistically significant superiority over existing methods. Final evaluations of QLOA-CM on engineering optimization problems and real-world challenges from CEC2020, alongside QLOA’s successful application in image segmentation tasks, further validate the effectiveness and generalization capabilities of the proposed approaches in solving complex optimization problems. The source code of the proposed algorithm is publicly available at https://github.com/MSNFV/QLOA-Q-Learning-based-Optimization-Algorithm.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"3110 - 3178"},"PeriodicalIF":4.4,"publicationDate":"2026-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903234","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Cross‑Scale Bioinspired Friction‑differential Stepping Positioning Platform Driven by Piezoelectric Bimorph Benders","authors":"Shupeng Wang,&nbsp;Xueliang Zhang,&nbsp;Xinbao Wang,&nbsp;Chaorui Jiang,&nbsp;Shihui Zhou,&nbsp;Zhihui Zhang","doi":"10.1007/s42235-026-00946-2","DOIUrl":"10.1007/s42235-026-00946-2","url":null,"abstract":"<div><p>While piezoelectric positioning platforms show significant potential in micro/nano actuation, their widespread adoption is often limited by inherent structural complexity. To address this, we proposed a cross-scale friction-differential stepping platform driven by only two piezoelectric bimorph benders. Utilizing a specially designed T-shaped structure, the platform achieved both linear and rotational motion through an innovative friction-differential principle. Following theoretical and dynamic simulation analyses, a prototype was fabricated. Experimental results demonstrate stable performance and high repeatability. For linear motion, the resolution was 2.853 <i>µ</i>m; at 150 V (no load), the maximum speed reached 3.306 mm/s (30 Hz), with a return deviation of 348.718 <i>µ</i>m (1 Hz). For rotational motion, the resolution was 0.063 mrad; the peak angular velocity attained 136.551 mrad/s (15 Hz), with a return deviation of 7.598 mrad. Furthermore, the platform exhibited a maximum payload-to-weight ratio of 63.49. This design offers a simple, redundancy-free 2-DOF solution, significantly reducing structural complexity compared to traditional platforms.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"3020 - 3036"},"PeriodicalIF":4.4,"publicationDate":"2026-08-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903271","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An Experimental Method for Dynamic Modeling of Hover-capable Tailless Flapping-wing Micro Air Vehicles","authors":"Heyu Cao,&nbsp;Jianghao Wu,&nbsp;Cheng Cheng,&nbsp;Yanlai Zhang,&nbsp;Long Chen,&nbsp;Peng Tang","doi":"10.1007/s42235-026-00957-z","DOIUrl":"10.1007/s42235-026-00957-z","url":null,"abstract":"<div><p>Despite the exceptional maneuverability of hover-capable tailless flapping-wing micro air vehicles, the inherent complexity of their unsteady aerodynamics poses significant challenges for developing precise dynamic models for controller design. To address this, this paper proposes ground-based damping pendulum experiments to construct a full damping dynamic model that incorporates parasitic drag damping. The parasitic drag damping is decoupled from flapping-induced damping by experimentally measuring the damping coefficients across three configurations: with wings and flapping, with wings and no flapping, and without wings. Quantitative analysis reveals that, compared with a model that considers only flapping-induced damping, the full damping dynamic model predicts the oscillation periods of longitudinal and lateral unstable modes decrease by 4.27% and 2.24%, respectively. The accuracy of the full damping model is validated by cascaded PID closed-loop control simulations; compared to flight test data, the root-mean-square errors for pitch and roll attitudes are only 1.78° and 2.78°, respectively. Furthermore, this study investigates the influence of wing geometry on aerodynamic derivatives. Results indicate that flapping-induced damping effects increase as the aspect ratio decreases. Finally, an experimentally corrected aerodynamic derivative estimation method is established. This study provides a feasible empirical modeling framework for the conceptual design and controller synthesis of FWMAVs.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"2823 - 2838"},"PeriodicalIF":4.4,"publicationDate":"2026-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EAL-Net: Evolutionary Algorithm-optimized Novel Lightweight Deep Learning Architecture with Multi-attention Modules and Orthogonal Softmax for Blood Cell Diagnosis","authors":"Omair Bilal,&nbsp;Yangfan Li,&nbsp;Sohaib Asif,&nbsp;Ming Zhao,&nbsp;Xiangmin Li","doi":"10.1007/s42235-026-00956-0","DOIUrl":"10.1007/s42235-026-00956-0","url":null,"abstract":"<div><p>Traditional diagnostic methods for Blood Cells (BCs) and Cervical Cells tend to erroneous analysis and inaccurate diagnoses due to the complexity involved in overlapping regions of cells. Deep Learning (DL) techniques through Convolutional Neural Networks (CNNs) have gained prominence in addressing these issues but encounter challenges as inefficient feature attention, and high computational processing power. To tackle these inefficiencies, this paper presents a novel computationally efficient lightweight architecture (EAL-Net) for BCs diagnosis, which is based on lightweight multi-attention modules and Orthogonal SoftMax Layer (OSL). OSL functionality precisely mitigates parameter co-adaptation to ensure orthogonality among weight vectors, thus refining feature learning processes. The developed EAL-Net architecture is equipped with novel Lightweight Attention Mechanism (LWAM) to selectively learn discriminative features and enabling more precise attention on diagnostic relevant regions. LWAM integrates Spatial-Attention Module (SAM) to focus on key regions, Spatial Self-Attention Module (SSAM) to capture long-range dependencies, and Category-Attention Module (CAM) to emphasize class-specific features. Collectively, this approach amplifies feature attention and strengthening spatial and channel-wise dependencies, significantly improving the performance of neural network architecture. Moreover, Genetic Algorithms (GA) efficiently search the optimal set of hyperparameters iteratively. GA is chosen for its efficacy in handling complex search spaces and non-linear optimization problems. Additionally, Grad-CAM and SHAP provide model insights by highlighting key features influencing predictions. EAL-Net outperforms pre-trained CNNs, transformer-based architectures, and existing methods across accuracy, efficiency, and memory use. With only 0.73 million parameters and 8.2 MB size, it achieves 96.11% accuracy on 8-class and 93.68% on 19-class BCs datasets. To validate the robustness of the proposed architecture across diverse medical imaging modalities, we evaluated EAL-Net on Cervical Cells (SIPaKMeD &amp; Mendeley_LBC), histopathological and chest X-ray datasets.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"3282 - 3311"},"PeriodicalIF":4.4,"publicationDate":"2026-08-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903554","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Artificial Neural Network Simulation of Radiative Hybrid Nanofluid Flow with Forchheimer Resistance on a Permeable Sheet","authors":"Raheela Razzaq,&nbsp;Zeeshan Khan,&nbsp;G. Dharmaiah,&nbsp;B. Hari Babu,&nbsp;Ho-Kin Tang,&nbsp;Cho Tung Yip","doi":"10.1007/s42235-026-00960-4","DOIUrl":"10.1007/s42235-026-00960-4","url":null,"abstract":"<div><p>This study presents a numerical simulation utilizing the Levenberg–Marquardt backpropagation (LMBP) algorithm to examine the radiative flow of hybridized cobalt ferrite <span>(left( {CoFe_{2} O_{4} } right))</span> and magnetite <span>(left( {Fe_{3} O_{4} } right))</span> nanofluids over a permeable stretching surface, with a focus on the Forchheimer effect. Hybrid nanofluids, which consist of nanoparticles with distinct thermophysical properties, are known to exhibit enhanced heat transfer performance compared to conventional single-phase fluids. The incorporation of radiation heat transfer and the Forchheimer effect into the fluid model offers a more comprehensive representation of complex heat transport processes relevant to industrial applications. A novel hybrid nanofluid composed of cobalt ferrite and magnetite nanoparticles is introduced, demonstrating a significant improvement in the thermal characteristics of the base fluid. A similarity transformation was applied to convert the governing partial differential equations into ordinary differential equations (ODEs). The impacts of various parameters on the flow field were investigated and visualized through graphical plots. Numerical solutions obtained via the BVP4C (boundary value problem 4C) method was compared with existing literature, showing excellent agreement. The findings contribute to the optimization of thermal management systems in advanced engineering and industrial processes. Additionally, the LMBP algorithm was validated through error analysis, regression metrics, adaptive control parameter evaluations, and iterative learning curve studies, demonstrating convergence, accuracy, and efficiency. Performance verification was conducted using regression plots, mean squared error (MSE), and error histograms, ensuring the robustness of the LMBP approach in solving the problem. In addition, the numerical solution is obtained by training the network with the LMPB algorithm by developing an accurate surrogate using 70%/10%/10% split for training/testing/validation. The LMPB algorithm obtained a best MSE of <span>(4 cdot 04419 times 10^{ - 9})</span> at epoch <span>(890)</span>, with errors closely grouped around zero and an overall regression coefficient <span>(R cong 1)</span> across the process demonstrating excellent agreement with the numerical targets.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"3335 - 3361"},"PeriodicalIF":4.4,"publicationDate":"2026-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903356","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bioinspired Quadrupedal Rock-Climbing Robot with Heterogeneous Fore–Hind Grippers and Knee Parallel Elastic Actuators","authors":"Peijin Zi,&nbsp;Yang Luo,&nbsp;Chang Wang,&nbsp;Bochao Song,&nbsp;Jiawei Chen,&nbsp;Tao Zhang,&nbsp;Kun Xu,&nbsp;Xilun Ding","doi":"10.1007/s42235-026-00963-1","DOIUrl":"10.1007/s42235-026-00963-1","url":null,"abstract":"<div><p>Robot rock climbing on natural terrain requires large and well-directed attachment forces on irregular surfaces, utilizing limited onboard energy. This paper presents the Multimodal Adaptive Rock-Climbing Robot Version 2 (MARCBot V2), a human-inspired quadrupedal climber with heterogeneous spiny grippers and minimalist knee Parallel Elastic Actuators (PEAs). The forelimb gripper uses underactuated fingers, Linearly Constrained Compliant Spines (LCCSs), and passive compliance. This design better aligns attachment-force directions with climbing loads, increasing tolerance to local surface geometry. The rubber-padded support mode enables walking on level ground. The hindlimb gripper utilizes a lightweight toe–heel layout, enabling toe-hook/heel-hook support to better resist slip and pitching during transitions and pitch changes. We designed a rubber-cord PEA using a Neo–Hookean model and multi-objective optimization. The PEA effectively reduces peak motor power and mitigates knee joint overload without adding complex mechanisms. Basalt tests report mean shear/normal forces of 62.2/25.9 N (maximum 76.2/31.3 N) for the forelimb. System demos demonstrate transitions from 0 to 70° and vertical climbing. PEA tests reveal a mean power reduction from 48.74 to 45.18 W (estimated 7.3% reduction in cost of transport).</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"2775 - 2800"},"PeriodicalIF":4.4,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903612","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Facial Expression Analysis, Drive Control Design, and Imitation Reproduction for Humanoid Robots","authors":"Manjia Su,&nbsp;Hongkai Chen,&nbsp;Haifei Zhu,&nbsp;Ruiwei Liu,&nbsp;Shichao Gu,&nbsp;Yisheng Guan","doi":"10.1007/s42235-026-00964-0","DOIUrl":"10.1007/s42235-026-00964-0","url":null,"abstract":"<div><p>Humanoid robots with realistic facial expression capabilities hold significant potential in medical therapy, education, and the arts by replicating human emotions, thereby enhancing adaptive Human–Robot Interaction (HRI). This work presents a systematic framework for developing a facial system for humanoid robots. Our contributions are fourfold: First, a data-driven method for facial skin drive point selection using the Cohn-Kanade dataset, analyzing landmark displacement patterns to differentiate static and dynamic facial regions. Second, a modular biomimetic facial system integrating six region-optimized actuation units (eyebrows, jaw, eyelids, nasal alar, eyes, mouth corners) with a total of 14 servo motors, a 3D-printed head-shell with static-dynamic partitioning, and anthropomorphic silicone skin fabricated using custom molds. Third, an expression imitation algorithm using 11 morphological geometric features and a neural inverse model (Multilayer Neural Network, MNN) trained on 3000 automatically collected servo-image pairs. Fourth, experimental validation demonstrating accurate expression replication through custom metrics (Mean Error Percentage of Command, MEPC; Mean Error Percentage of Discretization Command, MEPDC) and expression similarity quantification. Our system achieves MEPC of 0.107 and MEPDC of 0.114, outperforming Linear regression, Random Forest, and Support Vector Regression (SVR) models. This work establishes a foundational approach for developing expressive humanoid robot facial systems, enabling emotionally intelligent HRI.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"2971 - 2984"},"PeriodicalIF":4.4,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903611","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive Observation State Selection for a Quadruped Robot Under Varying Task Commands","authors":"Tianyu Zhang,&nbsp;Jieliang Zhao,&nbsp;Qun Niu,&nbsp;Xuemei Chen","doi":"10.1007/s42235-026-00954-2","DOIUrl":"10.1007/s42235-026-00954-2","url":null,"abstract":"<div><p>Reinforcement-learning-based controllers for lightweight robot control in structured, rather than complex, scenarios have received comparatively little attention. Lightweight control aims to achieve performance comparable to full observation-based control by relying on a minimal set of key observation factors. This study finds that different key observation factors become important under different commands. Therefore, the effectiveness of such methods still requires further enhancement. To better understand the influence of observation states on action generation and enhance lightweight control performance, this paper proposes a method based on an improved Integrated Gradients algorithm to quantify the importance of observation variables in quadruped robot control. By combining sampling and fitting techniques, the variation pattern of observation importance weights is derived with respect to different velocity commands, and the optimal observation subsets are identified under specific command conditions. To enable real-time switching of observation subsets during inference based on command input, a multi-encoder training framework is introduced. This framework encodes and learns multiple observation subsets using a shared replay buffer and a Multi-stage Encoder Switching Architecture. Simulation results demonstrate that different observation subsets are optimal for different command ranges. The proposed command-driven observation selection strategy improves the overall performance score by 5.33% compared with fixed Determined State Observation configurations. Real-world experiments validate similar trends, highlighting the potential of this approach for lightweight robot control and the exploration of biologically inspired locomotion mechanisms.</p></div>","PeriodicalId":614,"journal":{"name":"Journal of Bionic Engineering","volume":"23 5","pages":"2985 - 3000"},"PeriodicalIF":4.4,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}