Fuli Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Fuli Wang has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
Vishwanathan Mohan
School of Computer Science and Electronic Engineering, University of Essex, Colchester, UK
2 shared publications- Zuqiang Su
Chongqing University of Posts and Telecommunications
2 shared publications Ashutosh Tiwari
School of Mechanical, Aerospace and Civil Engineering, The University of Sheffield, Sheffield, UK
1 shared publication- Ashutosh Tiwari
School of Mechanical, Aerospace and Civil Engineering, The University of Sheffield, Sheffield, U.K.
1 shared publication Dongbing Gu
1 shared publication- Fazair Nizar Siraj
School of Mechanical, Aerospace and Civil Engineering, The University of Sheffield, Sheffield, U.K.
1 shared publication
Latest Publications
- 17 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (2) , 1874-1881
Physics-Informed Passive Motion Paradigm for Parallel Robots: A High-Precision Motor-Primitives Framework
Complex embodied systems, whether biological or robotic, must continuously generate goal-directed behaviors while preserving coherence between motor intention and physical feasibility. In parallel robots, this link between intention and mechanics becomes particularly challenging due to their nonlinear, over-constrained kinematics and the absence of intuitive motor primitives. This letter introduces a passive motion paradigm for parallel robots using self-supervised physics-informed neural networks, which reformulates motion generation as the dynamic unfolding of motor primitives driven by attractor fields in actuator space. Unlike traditional forward or optimization-based formulations, the framework integrates analytical kinematics with neural fields to ensure both physical consistency and adaptive motion generation. The method estimates the Jacobian matrix as a physically constrained neural field, merging analytical structure with data-driven learning to achieve robust and interpretable behavior without relying on iterative numerical solvers. Theoretical analysis, simulations, and physical experiments demonstrate the framework's accuracy, stability, and adaptability across different parallel mechanisms.
- 17 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 11 (2) , 1874-1881
Physics-Informed Passive Motion Paradigm for Parallel Robots: A High-Precision Motor-Primitives Framework
Complex embodied systems, whether biological or robotic, must continuously generate goal-directed behaviors while preserving coherence between motor intention and physical feasibility. In parallel robots, this link between intention and mechanics becomes particularly challenging due to their nonlinear, over-constrained kinematics and the absence of intuitive motor primitives. This letter introduces a passive motion paradigm for parallel robots using self-supervised physics-informed neural networks, which reformulates motion generation as the dynamic unfolding of motor primitives driven by attractor fields in actuator space. Unlike traditional forward or optimization-based formulations, the framework integrates analytical kinematics with neural fields to ensure both physical consistency and adaptive motion generation. The method estimates the Jacobian matrix as a physically constrained neural field, merging analytical structure with data-driven learning to achieve robust and interpretable behavior without relying on iterative numerical solvers. Theoretical analysis, simulations, and physical experiments demonstrate the framework's accuracy, stability, and adaptability across different parallel mechanisms.
- 21 April 2025
- Published by Cambridge University Press (CUP) in Robotica
- Vol. 43 (5) , 1766-1784
Passive motion paradigm implementation via deep neural networks: analysis and verification
In recent years, passive motion paradigms (PMPs), derived from the equilibrium point hypothesis and impedance control, have been utilised as manipulation methods for humanoid robots and robotic manipulators. These paradigms are typically achieved by creating a kinematic chain that enables the manipulator to perform goal-directed actions without explicitly solving the inverse kinematics. This approach leverages a kinematic model constructed through the training of artificial neural networks, aligning well with principles of cybernetics and cognitive computation by enabling adaptive and flexible control. Specifically, these networks model the relationship between joint angles and end-effector positions, facilitating the computation of the Jacobian matrix. Although this method does not require an accurate robot model, traditional neural networks often suffer from drawbacks such as overfitting and inefficient training, which can compromise the accuracy of the final PMP model. In this paper, we implement the method using a deep neural network and investigate the impact of activation functions and network depth on the performance of the kinematic model. Additionally, we propose a transfer learning approach to fine-tune the pre-trained model, enabling it to be transferred to other manipulator arms with different kinematic properties. Finally, we implement and evaluate the deep neural network-based PMP on the Universal Robots, comparing it with traditional kinematic controllers and assessing its physical interaction capabilities and accuracy.
- 21 April 2025
- Published by Cambridge University Press (CUP) in Robotica
- Vol. 43 (5) , 1766-1784
Passive motion paradigm implementation via deep neural networks: analysis and verification
In recent years, passive motion paradigms (PMPs), derived from the equilibrium point hypothesis and impedance control, have been utilised as manipulation methods for humanoid robots and robotic manipulators. These paradigms are typically achieved by creating a kinematic chain that enables the manipulator to perform goal-directed actions without explicitly solving the inverse kinematics. This approach leverages a kinematic model constructed through the training of artificial neural networks, aligning well with principles of cybernetics and cognitive computation by enabling adaptive and flexible control. Specifically, these networks model the relationship between joint angles and end-effector positions, facilitating the computation of the Jacobian matrix. Although this method does not require an accurate robot model, traditional neural networks often suffer from drawbacks such as overfitting and inefficient training, which can compromise the accuracy of the final PMP model. In this paper, we implement the method using a deep neural network and investigate the impact of activation functions and network depth on the performance of the kinematic model. Additionally, we propose a transfer learning approach to fine-tune the pre-trained model, enabling it to be transferred to other manipulator arms with different kinematic properties. Finally, we implement and evaluate the deep neural network-based PMP on the Universal Robots, comparing it with traditional kinematic controllers and assessing its physical interaction capabilities and accuracy.
- 1 November 2022
- Published by Elsevier in Biosystems Engineering
- Vol. 223, 103-114
An unsupervised automatic measurement of wheat spike dimensions in dense 3D point clouds for field application
- 1 November 2022
- Published by Elsevier in Biosystems Engineering
- Vol. 223, 103-114
An unsupervised automatic measurement of wheat spike dimensions in dense 3D point clouds for field application
- 16 June 2020
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 31 (9) , 095002
A fault diagnosis model based on singular value manifold features, optimized SVMs and multi-sensor information fusion
To achieve better fault diagnosis of rotating machinery, this paper presents a novel intelligent fault diagnosis model based on singular value manifold features (SVMF), optimized support vector machines (SVMs) and multi-sensor information fusion. Firstly, a new fault feature named SVMF is developed to better represent faults. SVMF is acquired by extracting manifold topology features of the singular spectrum. Compared with frequently-used fault features, the feature scale of SVMF is constant for variable rotating speed, and the extraction process of SVMF also has the effect of self-weighting. So SVMF has a better representation of faults. Then, to select optimal parameters for model training of SVMs, an improved fruit fly algorithm is proposed by introducing a guidance search mechanism and enhanced local search operation, and as a result both the convergence speed and accuracy are improved. Finally, the Dempster–Shafer evidence theory is introduced to fuse decision-level information from SVM models of multiple sensors. Information fusion eliminates the conflict of conclusions on fault diagnosis from multiple sensors, which leads to high robustness and accuracy of the fault diagnosis model. As a summary, the proposed method combines the advantages of SVMF in fault representation, SVMs in fault identification and the Dempster–Shafer evidence theory in information fusion, and as a result the proposed method will perform better at fault diagnosis. The proposed intelligent fault diagnosis model is subsequently applied to fault diagnosis of the gearbox. Experimental results show that the proposed diagnostic framework is versatile at detecting faults accurately.
- 16 June 2020
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 31 (9) , 095002
A fault diagnosis model based on singular value manifold features, optimized SVMs and multi-sensor information fusion
To achieve better fault diagnosis of rotating machinery, this paper presents a novel intelligent fault diagnosis model based on singular value manifold features (SVMF), optimized support vector machines (SVMs) and multi-sensor information fusion. Firstly, a new fault feature named SVMF is developed to better represent faults. SVMF is acquired by extracting manifold topology features of the singular spectrum. Compared with frequently-used fault features, the feature scale of SVMF is constant for variable rotating speed, and the extraction process of SVMF also has the effect of self-weighting. So SVMF has a better representation of faults. Then, to select optimal parameters for model training of SVMs, an improved fruit fly algorithm is proposed by introducing a guidance search mechanism and enhanced local search operation, and as a result both the convergence speed and accuracy are improved. Finally, the Dempster–Shafer evidence theory is introduced to fuse decision-level information from SVM models of multiple sensors. Information fusion eliminates the conflict of conclusions on fault diagnosis from multiple sensors, which leads to high robustness and accuracy of the fault diagnosis model. As a summary, the proposed method combines the advantages of SVMF in fault representation, SVMs in fault identification and the Dempster–Shafer evidence theory in information fusion, and as a result the proposed method will perform better at fault diagnosis. The proposed intelligent fault diagnosis model is subsequently applied to fault diagnosis of the gearbox. Experimental results show that the proposed diagnostic framework is versatile at detecting faults accurately.
- Open Access
- 1 May 2019
- Published by SAGE Publications in International Journal of Advanced Robotic Systems
- Vol. 16 (3)
A novel scheduling method for automated guided vehicles in workshop environments
At present, the commonly used task scheduling methods of automated guided vehicle do not fully consider the influence of power consumption, the workshop environment, and other factors, resulting in the disparity between scheduling methods and practical applications. This article contributes to filling this gap by modifying the model and algorithm that can meet the real-time application in the factory. First, a scheduling model is established according to both the number of depots and the automated guided vehicle’s battery consumption, so that the result of task allocation is more reasonable. Then, according to the area, distribution, shape characteristics of obstacles, and the number of depots contained in the environment, this article derives a new coefficient which is constructed as the weighted value of the distance between workstations to improve the robustness of the model. Finally, the modified genetic algorithm is used to obtain the scheduling results. The simulation results show the effectiveness and the rationality of the proposed method.
- 1 May 2019
- Published by SAGE Publications in International Journal of Advanced Robotic Systems
- Vol. 16 (3)
Open AccessA novel scheduling method for automated guided vehicles in workshop environments
At present, the commonly used task scheduling methods of automated guided vehicle do not fully consider the influence of power consumption, the workshop environment, and other factors, resulting in the disparity between scheduling methods and practical applications. This article contributes to filling this gap by modifying the model and algorithm that can meet the real-time application in the factory. First, a scheduling model is established according to both the number of depots and the automated guided vehicle’s battery consumption, so that the result of task allocation is more reasonable. Then, according to the area, distribution, shape characteristics of obstacles, and the number of depots contained in the environment, this article derives a new coefficient which is constructed as the weighted value of the distance between workstations to improve the robustness of the model. Finally, the modified genetic algorithm is used to obtain the scheduling results. The simulation results show the effectiveness and the rationality of the proposed method.