Dexin Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Dexin Wang has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Chunsheng Liu
School of Control Science and Engineering, Shandong University, Jinan 250061, China
7 shared publications - Faliang Chang
School of Control Science and Engineering, Shandong University, Jinan 250061, China
7 shared publications - Hengqiang Huan
School of Control Science and Engineering, Shandong University, Jinan, China
4 shared publications - Nanjun Li
School of Control Science and Engineering, Shandong University, Jinan, China
2 shared publications - Wenqian Wang
Pillar of Information Systems Technology and Design, Singapore University of Technology and Design, Singapore
2 shared publications - Dongmei Liu
Shandong Normal University
1 shared publication
Latest Publications
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 15123-15136
OAG-Pose: Orientation-Aware and Guided Category-Level Object Pose Estimation
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 15123-15136
OAG-Pose: Orientation-Aware and Guided Category-Level Object Pose Estimation
- 25 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (3) , 2867-2881
OSAD: Open-Set Supervised Anomaly Detection in Surveillance Videos Based on Margin Metric Learning
Open-set Supervised Anomaly Detection (OSAD) strategy seeks to detect novel anomalies that are unseen during training. However, existing OSAD works fail to learn a comprehensive margin that separates normal and anomalous samples in the feature space owing to the lack of restriction on abnormal distribution. To this end, we propose a new OSAD scheme based on Margin Metric Learning, which fully exploits the differences between normal and abnormal events in the latent feature dimension. The developed framework embodies three major components. First, video clips including normal events and restricted anomalies are input into an Attention-embedded Spatial Convolutional Network to extract the spatial feature sequences. Then, the spatial feature sequences are input into a Temporal Convolutional Siamese Network to further obtain the temporal features. Second, a novel Quadruplet Contrastive Loss is designed based on the spatial-temporal features of normal and abnormal video clips to conduct the margin learning, which enlarges the inter-class distances between anomalous and normal instances as well as reduces the intra-class distances inside anomalous instances and normal instances. This contributes to acquire compact normal and abnormal distributions. Finally, during the testing stage, a simplified metric distance-based anomaly detection strategy is proposed to calculate the anomaly score of testing clip. Extensive experiments and ablation studies on the Avenue and ShanghaiTech datasets demonstrate the effectiveness and efficiency of our proposed method for discovering unseen anomalous events via limited types of aware anomalies.
- 25 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 36 (3) , 2867-2881
OSAD: Open-Set Supervised Anomaly Detection in Surveillance Videos Based on Margin Metric Learning
Open-set Supervised Anomaly Detection (OSAD) strategy seeks to detect novel anomalies that are unseen during training. However, existing OSAD works fail to learn a comprehensive margin that separates normal and anomalous samples in the feature space owing to the lack of restriction on abnormal distribution. To this end, we propose a new OSAD scheme based on Margin Metric Learning, which fully exploits the differences between normal and abnormal events in the latent feature dimension. The developed framework embodies three major components. First, video clips including normal events and restricted anomalies are input into an Attention-embedded Spatial Convolutional Network to extract the spatial feature sequences. Then, the spatial feature sequences are input into a Temporal Convolutional Siamese Network to further obtain the temporal features. Second, a novel Quadruplet Contrastive Loss is designed based on the spatial-temporal features of normal and abnormal video clips to conduct the margin learning, which enlarges the inter-class distances between anomalous and normal instances as well as reduces the intra-class distances inside anomalous instances and normal instances. This contributes to acquire compact normal and abnormal distributions. Finally, during the testing stage, a simplified metric distance-based anomaly detection strategy is proposed to calculate the anomaly score of testing clip. Extensive experiments and ablation studies on the Avenue and ShanghaiTech datasets demonstrate the effectiveness and efficiency of our proposed method for discovering unseen anomalous events via limited types of aware anomalies.
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 41 (15523098) , 2086-2104
Hierarchical Diffusion Policy: Manipulation Trajectory Generation via Contact Guidance
Decision-making in robotics using denoising diffusion processes has increasingly become a hot research topic, but end-to-end policies perform poorly in tasks with rich contact and have limited interactivity. This paper proposes Hierarchical Diffusion Policy (HDP), a new robot manipulation policy of using contact points to guide the generation of robot trajectories. The policy is divided into two layers: the high-level policy predicts the contact for the robot's next object manipulation based on 3D information, while the low-level policy predicts the action sequence toward the high-level contact based on the latent variables of observation and contact. We represent both-level policies as conditional denoising diffusion processes, and combine behavioral cloning and Q-learning to optimize the low-level policy for accurately guiding actions towards contact. We benchmark Hierarchical Diffusion Policy across 6 different tasks and find that it significantly outperforms the existing state-of-the-art imitation learning method Diffusion Policy with an average improvement of 20.8%. We find that contact guidance yields significant improvements, including superior performance, greater interpretability, and stronger interactivity, especially on contact-rich tasks. To further unlock the potential of HDP, this paper proposes a set of key technical contributions including one-shot gradient optimization, trajectory augmentation, and prompt guidance, which improve the policy's optimization efficiency, spatial awareness, and interactivity respectively. Finally, real-world experiments verify that HDP can handle both rigid and deformable objects. Code, data and training details can be found at https://github.com/dexin-wang/Hierarchical-Diffusion-Policy/.
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 41 (15523098) , 2086-2104
Hierarchical Diffusion Policy: Manipulation Trajectory Generation via Contact Guidance
Decision-making in robotics using denoising diffusion processes has increasingly become a hot research topic, but end-to-end policies perform poorly in tasks with rich contact and have limited interactivity. This paper proposes Hierarchical Diffusion Policy (HDP), a new robot manipulation policy of using contact points to guide the generation of robot trajectories. The policy is divided into two layers: the high-level policy predicts the contact for the robot's next object manipulation based on 3D information, while the low-level policy predicts the action sequence toward the high-level contact based on the latent variables of observation and contact. We represent both-level policies as conditional denoising diffusion processes, and combine behavioral cloning and Q-learning to optimize the low-level policy for accurately guiding actions towards contact. We benchmark Hierarchical Diffusion Policy across 6 different tasks and find that it significantly outperforms the existing state-of-the-art imitation learning method Diffusion Policy with an average improvement of 20.8%. We find that contact guidance yields significant improvements, including superior performance, greater interpretability, and stronger interactivity, especially on contact-rich tasks. To further unlock the potential of HDP, this paper proposes a set of key technical contributions including one-shot gradient optimization, trajectory augmentation, and prompt guidance, which improve the policy's optimization efficiency, spatial awareness, and interactivity respectively. Finally, real-world experiments verify that HDP can handle both rigid and deformable objects. Code, data and training details can be found at https://github.com/dexin-wang/Hierarchical-Diffusion-Policy/.
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-16
SynergyPose: Synergy Enhanced Feature and Projection Decoupling for 6-D Pose Estimation
Estimating the 6D pose of texture-sparse objects is critical in vision-guided robotic manipulation and measurement. Existing single-RGB-based methods typically rely on multiple intermediate representations to improve accuracy, which increases computational and annotation costs. Alternatively, adopting a singular representation can mitigate these issues but it still suffers from feature scarcity, resulting in unreliable intermediate representation and pose estimation accuracy. Embracing these challenges, we propose a novel single-RGB-based framework named SynergyPose, which achieves accurate pose estimation of texture-sparse objects using only one intermediate representation. SynergyPose consists of three innovative components: Geometric Enhancer (GEer), Coordinate Projection Decoupling (CPD), and Consistency Constraint Loss (CCLoss). Specifically, we design a novel clustering branch based on feature aggregation and allocation within GEer, which synergistically extracts complementary features with a convolution branch to enhance feature representation. Based on the inter-projection coordinate consistency, we propose the CPD to capture the geometry of the object from different projection planes, which implicitly introduces multiple constraints to accurately calculate the dense correspondence without relying on other representations. Afterwards, CCLoss enforces coordinate consistency between the projection planes, explicitly introducing an additional constraint. Extensive experiments demonstrate the effectiveness of SynergyPose, which achieves competitive results on four BOP benchmark datasets.
- 5 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-16
SynergyPose: Synergy Enhanced Feature and Projection Decoupling for 6-D Pose Estimation
Estimating the 6D pose of texture-sparse objects is critical in vision-guided robotic manipulation and measurement. Existing single-RGB-based methods typically rely on multiple intermediate representations to improve accuracy, which increases computational and annotation costs. Alternatively, adopting a singular representation can mitigate these issues but it still suffers from feature scarcity, resulting in unreliable intermediate representation and pose estimation accuracy. Embracing these challenges, we propose a novel single-RGB-based framework named SynergyPose, which achieves accurate pose estimation of texture-sparse objects using only one intermediate representation. SynergyPose consists of three innovative components: Geometric Enhancer (GEer), Coordinate Projection Decoupling (CPD), and Consistency Constraint Loss (CCLoss). Specifically, we design a novel clustering branch based on feature aggregation and allocation within GEer, which synergistically extracts complementary features with a convolution branch to enhance feature representation. Based on the inter-projection coordinate consistency, we propose the CPD to capture the geometry of the object from different projection planes, which implicitly introduces multiple constraints to accurately calculate the dense correspondence without relying on other representations. Afterwards, CCLoss enforces coordinate consistency between the projection planes, explicitly introducing an additional constraint. Extensive experiments demonstrate the effectiveness of SynergyPose, which achieves competitive results on four BOP benchmark datasets.
- 11 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (7) , 6712-6719
Multi-Stage Reinforcement Learning for Non-Prehensile Manipulation
Manipulating objects without grasping them facilitates complex tasks, known as non-prehensile manipulation. Most previous methods are limited to learning a single skill to manipulate objects with primitive shapes and are unserviceable for flexible object manipulation that requires a combination of multiple skills. We explore skill-unconstrained non-prehensile manipulation and propose a Multi-stage Reinforcement Learning for Non-prehensile Manipulation ( MRLNM ), which calculates an intermediate state between the initial and goal states and divides the task into multiple stages for sequential learning. At each stage, the policy takes the desired 6-DOF object pose as the goal, and proposes a spatially-continuous action, allowing the robot to explore arbitrary skills to accomplish the task. To handle objects with different shapes, we propose a State-Goal Fusion Representation (SGF-Representation) to represent observations and goals as point clouds with motion, which improves the policy's perception of scene layout and task goal. To improve sample efficiency, we propose a Spatially-Reachable Distance Metric (SR-Distance) to approximately measure the shortest distance between two points without intersecting the scene. We evaluate MRLNM on an occluded grasping task which aims to grasp the object in initially occluded configurations. MRLNM demonstrates generalization to unseen objects with shapes outside the training distribution and can be transferred to the real world with zero-shot transfer, achieving a 95% success rate.
- 11 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (7) , 6712-6719
Multi-Stage Reinforcement Learning for Non-Prehensile Manipulation
Manipulating objects without grasping them facilitates complex tasks, known as non-prehensile manipulation. Most previous methods are limited to learning a single skill to manipulate objects with primitive shapes and are unserviceable for flexible object manipulation that requires a combination of multiple skills. We explore skill-unconstrained non-prehensile manipulation and propose a Multi-stage Reinforcement Learning for Non-prehensile Manipulation ( MRLNM ), which calculates an intermediate state between the initial and goal states and divides the task into multiple stages for sequential learning. At each stage, the policy takes the desired 6-DOF object pose as the goal, and proposes a spatially-continuous action, allowing the robot to explore arbitrary skills to accomplish the task. To handle objects with different shapes, we propose a State-Goal Fusion Representation (SGF-Representation) to represent observations and goals as point clouds with motion, which improves the policy's perception of scene layout and task goal. To improve sample efficiency, we propose a Spatially-Reachable Distance Metric (SR-Distance) to approximately measure the shortest distance between two points without intersecting the scene. We evaluate MRLNM on an occluded grasping task which aims to grasp the object in initially occluded configurations. MRLNM demonstrates generalization to unseen objects with shapes outside the training distribution and can be transferred to the real world with zero-shot transfer, achieving a 95% success rate.
- 15 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 71 (7) , 7388-7399
On-Policy and Pixel-Level Grasping Across the Gap Between Simulation and Reality
Grasp detection in cluttered scenes is a very challenging task for robots. Generating synthetic grasping data is a popular way to train and test grasp methods, as is Dex-Net; yet, these methods sample training grasps on 3-D synthetic object models, but evaluate at images or point clouds with different sample distributions, which reduces performance due to covariate shift and sparse grasp labels. To solve existing problems, we propose a novel on-policy grasp detection method for parallel grippers, which can train and test on the approximate distribution with dense pixel-level grasp labels generated on RGB-D images. An Orthographic-Depth Grasp Generation (ODG-Generation) method is proposed to generate an orthographic depth image through a new imaging model of projecting points in orthographic; then this method generates multiple candidate grasps for each pixel and obtains robust positive grasps through flatness detection, force-closure metric and collision detection. Then, a comprehensive Pixel-Level Grasp Pose Dataset (PLGP-Dataset) is constructed, which is the first pixel-level grasp dataset, with the on-policy distribution. Lastly, we build a grasp detection network with a novel data augmentation process for imbalance training. Experiments show that our on-policy method can partially overcome the gap between simulation and reality, and achieves the best performance.
- 15 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 71 (7) , 7388-7399
On-Policy and Pixel-Level Grasping Across the Gap Between Simulation and Reality
Grasp detection in cluttered scenes is a very challenging task for robots. Generating synthetic grasping data is a popular way to train and test grasp methods, as is Dex-Net; yet, these methods sample training grasps on 3-D synthetic object models, but evaluate at images or point clouds with different sample distributions, which reduces performance due to covariate shift and sparse grasp labels. To solve existing problems, we propose a novel on-policy grasp detection method for parallel grippers, which can train and test on the approximate distribution with dense pixel-level grasp labels generated on RGB-D images. An Orthographic-Depth Grasp Generation (ODG-Generation) method is proposed to generate an orthographic depth image through a new imaging model of projecting points in orthographic; then this method generates multiple candidate grasps for each pixel and obtains robust positive grasps through flatness detection, force-closure metric and collision detection. Then, a comprehensive Pixel-Level Grasp Pose Dataset (PLGP-Dataset) is constructed, which is the first pixel-level grasp dataset, with the on-policy distribution. Lastly, we build a grasp detection network with a novel data augmentation process for imbalance training. Experiments show that our on-policy method can partially overcome the gap between simulation and reality, and achieves the best performance.
- 26 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 69 (11) , 11611-11621
High-Performance Pixel-Level Grasp Detection Based on Adaptive Grasping and Grasp-Aware Network
Machine vision-based planar grasping detection is challenging due to uncertainty about object shape, pose, size, etc. Previous methods mostly focus on predicting discrete gripper configurations, and may miss some ground-truth grasp postures. In this article, a pixel-level grasp detection method is proposed, which uses deep neural network to predict pixel-level gripper configurations on RGB images. First, a novel oriented arrow representation model (OAR-model) is introduced to represent the gripper configuration of parallel-jaw and three-fingered gripper, which can partly improve the applicability to different grippers. Then, the adaptive grasping attribute model is proposed to adaptively represent the grasping attribute of objects, for resolving angle conflicts in training and simplifying pixel-level labeling. Lastly, the adaptive feature fusion and grasp-aware network (AFFGA-Net) is proposed to predict pixel-level OAR-models on RGB images. AFFGA-Net improves the robustness in unstructured scenarios by using hybrid atrous spatial pyramid and adaptive decoder connected in sequence. On the public Cornell dataset and actual objects, our structure achieves 99.09% and 98.0% grasp detection accuracy, respectively. In over 2400 robotic grasp trials, our structure achieves an average success rate of 98.77% in single-object scenarios and 93.69% in cluttered scenarios. Moreover, AFFGA-Net completes a grasp detection pipeline within 15 ms.
- 26 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Electronics
- Vol. 69 (11) , 11611-11621
High-Performance Pixel-Level Grasp Detection Based on Adaptive Grasping and Grasp-Aware Network
Machine vision-based planar grasping detection is challenging due to uncertainty about object shape, pose, size, etc. Previous methods mostly focus on predicting discrete gripper configurations, and may miss some ground-truth grasp postures. In this article, a pixel-level grasp detection method is proposed, which uses deep neural network to predict pixel-level gripper configurations on RGB images. First, a novel oriented arrow representation model (OAR-model) is introduced to represent the gripper configuration of parallel-jaw and three-fingered gripper, which can partly improve the applicability to different grippers. Then, the adaptive grasping attribute model is proposed to adaptively represent the grasping attribute of objects, for resolving angle conflicts in training and simplifying pixel-level labeling. Lastly, the adaptive feature fusion and grasp-aware network (AFFGA-Net) is proposed to predict pixel-level OAR-models on RGB images. AFFGA-Net improves the robustness in unstructured scenarios by using hybrid atrous spatial pyramid and adaptive decoder connected in sequence. On the public Cornell dataset and actual objects, our structure achieves 99.09% and 98.0% grasp detection accuracy, respectively. In over 2400 robotic grasp trials, our structure achieves an average success rate of 98.77% in single-object scenarios and 93.69% in cluttered scenarios. Moreover, AFFGA-Net completes a grasp detection pipeline within 15 ms.
- Open Access
- 12 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
High-performance Pixel-level Grasp Detection based on Adaptive Grasping and Grasp-aware Network
A pixel-level grasp detection method. https://youtu.be/ccA1jkkbBJA
- 12 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessHigh-performance Pixel-level Grasp Detection based on Adaptive Grasping and Grasp-aware Network
A pixel-level grasp detection method. https://youtu.be/ccA1jkkbBJA