Yichen Fu

Yichen Fu

Auto-generated profile, based on publicly available information

Yichen Fu

School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 January 2022 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
YFYichen Fu
Yichen Fu
junfeng Fanjunfeng Fan
junfeng Fan
SXShiyu Xing
Shiyu Xing
ZWZhe Wang
Zhe Wang
FJFengshui Jing
Fengshui Jing
MTMin Tan
Min Tan
Cabin pose measurement is one of the key procedures in the assembly and docking process of large cabins, which provides important feedback information for the subsequent docking control system. As the basis of cabin pose measurement, the accuracy and robustness of cabin assembly image segmentation are particularly important. However, traditional image segmentation method based on RGB sensor is extremely susceptible to interference from the external environment, which greatly weakens the recognition effect. In this article, an image segmentation method of cabin assembly scene based on improved red-green-blue-depth (RGB-D) Mask R-CNN is proposed, and its network structure is designed to be able to specifically process four-channel images. The method can accurately extract the corresponding area of the cabin under complex and severe environmental disturbances, with high robustness and generalization capability. Meanwhile, the excellence of deep learning segmentation algorithms with depth channel information input is highlighted. In experiments, improved classic segmentation network U-Net, SegNet, pyramid scene parsing network (PSPNet), and Deeplab-v3 based on RGB-D were constructed as control, and these models were tested and evaluated on the enhanced test sets to verify their segmentation accuracy and robustness performance. Comparing experiments fully demonstrate the superiority of the segmentation network model of RGB-D four-channel input over RGB input. At the same time, vision system using the proposed Mask R-CNN algorithm based on RGB-D has the best cabin segmentation accuracy, robustness, and generalization capability, which has practical significance for industrial applications.

25 August 2023 in IEEE Transactions on Industrial Electronics
Verified
Journal Article
Closed access
YFYichen Fu
Yichen Fu
junfeng Fanjunfeng Fan
junfeng Fan
FJFengshui Jing
Fengshui Jing
MTMin Tan
Min Tan
Fringe projection profilometry is one of the main methods for active 3-D measurement, but practical problems, such as how to obtain a complete point cloud by overcoming the effects of high surface reflectivity of objects, remain to be solved. So, in this article, we propose a set of hardware devices and region adaptive structured light algorithms. First, we introduce the surface structured light measurement system adopted in this article. Second, based on the accurate optical modeling of the measurement scene, the principles for generating the adaptive optimal projection brightness are derived in detail and the process is accelerated by exploiting the speed of neural network. Meanwhile, we creatively use the chain code combined with the M-estimator sample consensus method to find the homography from the saturated region of the camera plane to the corresponding region of the projector plane for generating fringe images with adaptive brightness. Furthermore, we chose the more robust line-shifting code and binarization method for subpixel 3-D reconstruction. Experiments have shown that our work is efficient, superior, and accurate in measuring objects with high reflective surface properties.

26 September 2025 in IEEE Transactions on Automation Science and Engineering
Verified
Journal Article
Closed access
YMYunkai Ma
Yunkai Ma
junfeng Fanjunfeng Fan
junfeng Fan
JHJun Hou
Jun Hou
YFYichen Fu
Yichen Fu
RTRui Tao
Rui Tao
FJFengshui Jing
Fengshui Jing
Robotic welding is an essential technology in industrial production. Various sensors are integrated into robotic welding systems to enhance welding efficiency and quality. Due to their advantages of non-contact measurement, high information capacity, and high accuracy, vision sensors play an increasingly important role in robotic autonomous welding. Therefore, the latest vision-based autonomous robotic arc welding technologies are comprehensively summarized, providing valuable reference and support for researchers. First, the research progress of visual sensing technology for welding robots is outlined, and its advantages and disadvantages are analyzed. Subsequently, the key technologies of robotic autonomous welding are introduced, covering weld type identification, initial point guidance, welding path generation, welding parameter planning, feature point extraction, seam tracking, welding posture adjustment, and welding quality control. Finally, the limitations existing in current research are summarized, and the future research directions of autonomous robotic welding are prospected.

31 May 2024 in IEEE Transactions on Automation Science and Engineering
Verified
Journal Article
Closed access
RTRui Tao
Rui Tao
FJFengshui Jing
Fengshui Jing
JHJun Hou
Jun Hou
SXShiyu Xing
Shiyu Xing
YFYichen Fu
Yichen Fu
MTMin Tan
Min Tan
The Five Hundred Meter Spherical Radio Telescope (FAST) is the largest spherical radio telescope in the world, and there are more than 2,000 prism targets distributed on its reflector that require regular maintenance. These prism targets are screwed into the corresponding threaded holes by target bolts. At present, manual maintenance is mainly used, which is inefficient and unsafe. To address this issue, we develop an autonomous prism target maintenance robotic system, called APTMRS. This system integrates a mobile robot with an assembly robot and a collaborative robot, enabling it to move and perform the prism target replacement task on the reflector. Its technical development is threefold. First, a robotic end-effector that can be used for screwing the non-standard target bolt from the side is designed. Second, a novel pose measurement framework, which incorporates a feature point extraction and matching method, is utilized to measure the pose of the prism target. Third, the manipulation policies for the challenging steps in the workflow are presented, including picking and placing, initial thread mating, and final tightening. Independent experiments indicate that the performances of these methods are satisfactory. Moreover, tests in both simulated and real FAST scenarios confirm the effectiveness of APTMRS in achieving autonomous prism target maintenance. Note to Practitioners —Due to its excellent versatility, threaded connections have been widely adopted in a multitude of scenarios. Threads are not only present on standard bolts and nuts but also on a myriad of non-standard components. The robotic system proposed in this paper offers a solution for assembling and disassembling target bolts for FAST. Specifically, the scenario applied in this paper covers several extremely challenging problems in autonomous threaded assembly. Firstly, the object of assembly is a non-standard bolt, with obstacles above it, rendering traditional tools inapplicable. Secondly, the surface of the object is weakly textured and highly reflective. These pose significant challenges for the vision system to accurately acquire its pose. Lastly, the assembly environment is harsh. The threaded holes reside on an incline that is hard to access, with a maximum slope of $56^\circ$ . To address these challenges, this paper presents a robotic system supplemented by three techniques including a novel robotic end-effector, a new pose measurement framework, and manipulation policies. The experimental results indicate that this robotic system is capable of autonomously disassembling and assembling non-standard target bolts. We explore the potential for generalizing these designs and methods to other scenarios, with the hope of contributing new insights to the development of threaded assembly systems.

19 September 2024 in IEEE/ASME Transactions on Mechatronics
Verified
Journal Article
Closed access
YFYichen Fu
Yichen Fu
junfeng Fanjunfeng Fan
junfeng Fan
YMYunkai Ma
Yunkai Ma
FJFengshui Jing
Fengshui Jing
MTMin Tan
Min Tan
Point clouds of reflective objects are often incomplete in fringe projection structured light 3-D measurement. Traditional solutions are always computationally inefficient, whereas the speedy forward propagation of neural networks can overcome this drawback. So, in this article, we propose a novel deep learning method capable of adaptively controlling the projection brightness in specific regions with a high dynamic range for complete point cloud acquisition. First, we describe the structured light system (SLS) and the physical model used. Second, we construct a convolutional neural network (CNN) incorporating physical prior knowledge, which enables full-process high-resolution multichannel image feature extraction and directly outputs pixel-level optimal projection brightness. Meanwhile, we propose a simulated dataset generation method based on stochastic reflectivity together with a network training approach to solve network training difficulties. Further, the robust fringe encoding and decoding methods are presented. Sufficient comparative experiments show that our CNN-based SLS can achieve more accurate brightness adjustment with fewer projections, higher computational efficiency, and applicability.

15 August 2025 in IEEE/ASME Transactions on Mechatronics
Verified
Journal Article
Closed access
RTRui Tao
Rui Tao
YMYunkai Ma
Yunkai Ma
junfeng Fanjunfeng Fan
junfeng Fan
YFYichen Fu
Yichen Fu
FJFengshui Jing
Fengshui Jing
MTMin Tan
Min Tan
Robotic threaded assembly is crucial in automation, yet existing methods struggle with nonstandard and structurally diverse objects. To address this issue, we develop a grasping-based robotic threaded assembly system. The system is built around a custom-designed grasp-based screwing tool, and leverages point cloud input to generate grasping poses and motions. The system also allows autonomous fingertip replacement, enabling continuous assembly of various objects. This article mainly addresses two key challenges caused by uncertainty in the assembly process: Feeding and screwing. Notably, the feeding process is essentially a handover task. First, we propose an object handover method that integrates gradient-descent search with taboo search, ensuring robust transfer of the object to the tool’s grasping center, while aligning the screwing axis with the object’s threaded axis. Second, we introduce a synchronized position and orientation search strategy, followed by fine-tuning of the orientation, to enable precise and stable threaded assembly. The experimental results validate the effectiveness of the proposed methods. The system successfully achieves continuous assembly of objects with diverse shapes, demonstrating its adaptability and robustness.