Selected work

Representative Papers

Affordance-Based Manipulation Planning with Text Goals and Sim-to-Real Generalisation via Real-to-Sim Image Conversion

Jul 12, 2026

This work addresses the challenge of enabling robots to reliably plan and execute tasks from natural language instructions in complex, dynamic environments with occlusions, while achieving effective sim-to-real transfer. To this end, the authors propose a planning framework that integrates functional affordance recognition with visual action-effect prediction, leveraging visual forward reasoning to anticipate future states. A multimodal text-image matching module is introduced to evaluate the consistency between candidate action sequences and the linguistic goal. Furthermore, a real-to-sim image stylization mechanism is designed to enhance perceptual robustness in real-world settings. Experimental results demonstrate that the proposed approach successfully accomplishes challenging manipulation tasks on both simulated and physical robot platforms, significantly improving language-conditioned generalization from simulation to reality.

0 citationsRead paper

Estimating Velocity of Spheres from Rolling-Shutter Image(s)

Jun 30, 2026

This study addresses the challenge of accurately estimating both 3D translational and angular velocities of a rapidly moving sphere from a single rolling-shutter image. To this end, the authors propose a two-stage decoupled optimization framework: first, a highly discernible spherical pattern is designed to encode rolling-shutter distortions as temporally modulated signals; second, a geometric consistency back-projection model—requiring no feature correspondences—is formulated to separately recover translational and rotational velocities. This work is the first to effectively exploit rolling-shutter distortion as a source of motion information in a single frame, thereby overcoming the observability limitations of textureless spheres in extreme high-speed scenarios. Experiments demonstrate that the proposed method achieves robust and accurate velocity estimation on both synthetic and real-world datasets.

0 citationsRead paper

SI-Diff: A Framework for Learning Search and High-Precision Insertion with a Force-Domain Diffusion Policy

May 12, 2026

This work addresses the challenge of unifying search and high-precision insertion behaviors in contact-rich assembly tasks, where relative pose uncertainty complicates joint modeling. To this end, the authors propose SI-Diff, a framework that leverages a force-domain diffusion strategy to jointly learn both behaviors and introduces a novel mode-conditioning mechanism enabling a single policy to adaptively switch between search and insertion modes. By integrating tactile and end-effector velocity observations, teacher–student imitation learning, and a new search teacher policy that generates diverse trajectories, SI-Diff significantly enhances generalization. Compared to the TacDiffusion baseline, it improves lateral (x–y) misalignment tolerance from 2 mm to 5 mm and demonstrates strong zero-shot transfer performance on unseen object geometries.

0 citationsRead paper

Perception-Control Coupled Visual Servoing for Textureless Objects Using Keypoint-Based EKF

Feb 06, 2026

This work addresses the instability and low accuracy of visual servoing for textureless objects under adverse visual conditions such as occlusion, where conventional feature-based methods fail due to insufficient visual cues. To overcome this limitation, the authors propose a tightly coupled perception-control closed-loop visual servoing framework. The approach integrates learned keypoint detection with an Extended Kalman Filter (EKF) to jointly estimate the object’s 6D pose, which drives a pose-based visual servoing (PBVS) controller. Camera motion feedback is incorporated to enhance the robustness of keypoint tracking. Furthermore, a novel uncertainty-aware probabilistic control law is introduced to enable safe and precise manipulation of textureless objects. Real-world robotic experiments demonstrate that the proposed method significantly outperforms traditional visual servoing techniques in both pose estimation accuracy and grasping success rate.

0 citationsRead paper

Empirical Investigation of the Impact of Phase Information on Fault Diagnosis of Rotating Machinery

Dec 17, 2025

In rotating machinery fault diagnosis, phase information has long been neglected or discarded, with its practical impact lacking systematic empirical validation. This paper systematically reveals the critical role of phase alignment in multi-axis vibration-based diagnosis. We propose two phase-aware preprocessing strategies: (1) axis-wise independent phase zeroing and (2) single-axis reference-based phase alignment—which preserves inter-axis spatial phase relationships and significantly enhances robustness. Leveraging a custom-synchronized tri-axial rotor vibration dataset, we integrate time-domain phase calibration with spectrum-domain phase-sensitive modeling within a two-stage deep learning framework across six architectures. Experimental results show that the single-axis reference alignment achieves 96.2% classification accuracy—outperforming the baseline by 5.4%. Axis-wise independent alignment also yields consistent gains across diverse models (e.g., +2.7% for Transformer), demonstrating the universal utility of phase information in vibration-based fault diagnosis.

0 citationsRead paper

Recent publications

Latest Papers

Affordance-Based Manipulation Planning with Text Goals and Sim-to-Real Generalisation via Real-to-Sim Image Conversion

Jul 12, 2026

This work addresses the challenge of enabling robots to reliably plan and execute tasks from natural language instructions in complex, dynamic environments with occlusions, while achieving effective sim-to-real transfer. To this end, the authors propose a planning framework that integrates functional affordance recognition with visual action-effect prediction, leveraging visual forward reasoning to anticipate future states. A multimodal text-image matching module is introduced to evaluate the consistency between candidate action sequences and the linguistic goal. Furthermore, a real-to-sim image stylization mechanism is designed to enhance perceptual robustness in real-world settings. Experimental results demonstrate that the proposed approach successfully accomplishes challenging manipulation tasks on both simulated and physical robot platforms, significantly improving language-conditioned generalization from simulation to reality.

0 citationsRead paper

Estimating Velocity of Spheres from Rolling-Shutter Image(s)

Jun 30, 2026

This study addresses the challenge of accurately estimating both 3D translational and angular velocities of a rapidly moving sphere from a single rolling-shutter image. To this end, the authors propose a two-stage decoupled optimization framework: first, a highly discernible spherical pattern is designed to encode rolling-shutter distortions as temporally modulated signals; second, a geometric consistency back-projection model—requiring no feature correspondences—is formulated to separately recover translational and rotational velocities. This work is the first to effectively exploit rolling-shutter distortion as a source of motion information in a single frame, thereby overcoming the observability limitations of textureless spheres in extreme high-speed scenarios. Experiments demonstrate that the proposed method achieves robust and accurate velocity estimation on both synthetic and real-world datasets.

0 citationsRead paper

SI-Diff: A Framework for Learning Search and High-Precision Insertion with a Force-Domain Diffusion Policy

May 12, 2026

This work addresses the challenge of unifying search and high-precision insertion behaviors in contact-rich assembly tasks, where relative pose uncertainty complicates joint modeling. To this end, the authors propose SI-Diff, a framework that leverages a force-domain diffusion strategy to jointly learn both behaviors and introduces a novel mode-conditioning mechanism enabling a single policy to adaptively switch between search and insertion modes. By integrating tactile and end-effector velocity observations, teacher–student imitation learning, and a new search teacher policy that generates diverse trajectories, SI-Diff significantly enhances generalization. Compared to the TacDiffusion baseline, it improves lateral (x–y) misalignment tolerance from 2 mm to 5 mm and demonstrates strong zero-shot transfer performance on unseen object geometries.

0 citationsRead paper

Perception-Control Coupled Visual Servoing for Textureless Objects Using Keypoint-Based EKF

Feb 06, 2026

This work addresses the instability and low accuracy of visual servoing for textureless objects under adverse visual conditions such as occlusion, where conventional feature-based methods fail due to insufficient visual cues. To overcome this limitation, the authors propose a tightly coupled perception-control closed-loop visual servoing framework. The approach integrates learned keypoint detection with an Extended Kalman Filter (EKF) to jointly estimate the object’s 6D pose, which drives a pose-based visual servoing (PBVS) controller. Camera motion feedback is incorporated to enhance the robustness of keypoint tracking. Furthermore, a novel uncertainty-aware probabilistic control law is introduced to enable safe and precise manipulation of textureless objects. Real-world robotic experiments demonstrate that the proposed method significantly outperforms traditional visual servoing techniques in both pose estimation accuracy and grasping success rate.

0 citationsRead paper

Empirical Investigation of the Impact of Phase Information on Fault Diagnosis of Rotating Machinery

Dec 17, 2025

In rotating machinery fault diagnosis, phase information has long been neglected or discarded, with its practical impact lacking systematic empirical validation. This paper systematically reveals the critical role of phase alignment in multi-axis vibration-based diagnosis. We propose two phase-aware preprocessing strategies: (1) axis-wise independent phase zeroing and (2) single-axis reference-based phase alignment—which preserves inter-axis spatial phase relationships and significantly enhances robustness. Leveraging a custom-synchronized tri-axial rotor vibration dataset, we integrate time-domain phase calibration with spectrum-domain phase-sensitive modeling within a two-stage deep learning framework across six architectures. Experimental results show that the single-axis reference alignment achieves 96.2% classification accuracy—outperforming the baseline by 5.4%. Axis-wise independent alignment also yields consistent gains across diverse models (e.g., +2.7% for Transformer), demonstrating the universal utility of phase information in vibration-based fault diagnosis.

0 citationsRead paper