A5028503217- Robot Manipulation and Learning
- Tactile and Sensory Interactions
- Muscle activation and electromyography studies
- Advanced Sensor and Energy Harvesting Materials
- Robotics and Automated Systems
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Reinforcement Learning in Robotics
- Soft Robotics and Applications
- Human Pose and Action Recognition
- Speech and dialogue systems
- Adversarial Robustness in Machine Learning
- Manufacturing Process and Optimization
- EEG and Brain-Computer Interfaces
- Topic Modeling
- Industrial Vision Systems and Defect Detection
- Machine Learning in Materials Science
- Mechanics and Biomechanics Studies
Shandong Institute of Automation
2020-2025
Chinese Academy of Sciences
2020-2025
Institute of Automation
2024
University of Chinese Academy of Sciences
2020-2023
In-hand object localization and manipulation has always been a challenging task in robotic community. In this article, we address problem by vision-based tactile sensing with high-spatial resolution. Specifically, design novel sensor based on stereo vision, named GelStereo, which can perceive point cloud resolution ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$<\!1$</tex-math></inline-formula> mm). A...
Predicting whether a particular grasp will succeed is critical to performing stable grasping and manipulating tasks. Robots need combine vision touch as humans do accomplish this prediction. The primary problem be solved in process how learn effective visual-tactile fusion features. In letter, we propose novel Visual-Tactile Fusion learning method based on the Self-Attention mechanism (VTFSA) address problem. We compare proposed with traditional methods two public multimodal datasets,...
Humans can quickly determine the force required to grasp a deformable object prevent its sliding or excessive deformation through vision and touch, which is still challenging task for robots. To address this issue, we propose novel 3D convolution-based visual-tactile fusion deep neural network (C3D-VTFN) evaluate state of various objects in paper. Specifically, divide states into three categories sliding, appropriate excessive. Also, dataset training testing proposed built by extensive...
Next‐generation robots are being designed to function autonomously in complex and unstructured environments. In particular, based on the real‐time measurement differentiation of normal pressure shear force, can be equipped with capabilities damage‐free grasp within minimum force limits, as well dexterous operation through surface roughness slip information. Herein, a flexible tactile sensor small cylinder protrusion four arc‐shaped protrusions is developed. Due its center symmetry...
Adapting the mastered manipulation skill to novel objects is still challenging for robots. Recent works have attempted endow robot with ability adapt unseen tasks by leveraging meta-learning. However, these methods are data-hungry in training phase, which limits their application real world. In this paper, we propose Meta-Residual Policy Learning (MRPL) reduce cost of policy learning and adaptation. During meta-training, MRPL accelerates process focusing on residual task-shared knowledge...
Unknown surface material classification (SMC) can inform a robot about properties, enabling it to interact with environments appropriately. Recent research has leveraged multimodal data using deep learning improve the performance of SMC. In this article, we present model, temporal convolutional neural network (MTCNN), which integrates energy spectrum, dilated convolutions, and sequence poolings into unified architecture. The proposed model learn representations from auditory multitactile...
Slip detection plays a vital role in robotic dexterous grasping and manipulation, it has long been challenging problem the community. Different from traditional tactile perception-based methods, we propose Generalized Visual-Tactile Transformer (GVT-Transformer) network to detect slip based on visual spatiotemporal sequences. The main novelty of GVT-Transformer is its ability address unaligned vision data various formats captured by sensors. Furthermore, train test our proposed public...
Tool usage is critical for enabling robots to complete challenging tasks that exceed their innate capabilities. Task-oriented grasp and manipulation are two primitive actions in tool tasks. In this paper, we present an end-to-end framework jointly inferring through self-supervision, which can guide We formulate as oriented keypoint representations so the existing pose-based policies be easily used achieve To address low task completion rates propose a novel technique based on...
Humans can quickly determine the force required to grasp a deformable object prevent its sliding or excessive deformation through vision and touch, which is still challenging task for robots. To address this issue, we propose novel 3D convolution-based visual-tactile fusion deep neural network (C3D-VTFN) evaluate state of various objects in paper. Specifically, divide states into three categories sliding, appropriate excessive. Also, dataset training testing proposed built by extensive...
Endowing robots with tool manipulation skills helps them accomplish challenging tasks. While manipulate tools to achieve goals, the alignment of and targets is a noise-sensitive contact-rich task. However, it difficult access accurate pose target. When there unknown noise in observations, reinforcement learning can't be sure perform well. In this paper, we define easier-to-obtain task-related information as anchor introduce method based on information, which can well when observations...
Zero-shot imitation learning has demonstrated its superiority to learn complex robotic tasks with less human participation. Recent studies show convincing performance under the condition that robot follows demonstration strictly by learned inverse model. However, these methods are difficult achieve satisfactory in when is suboptimal, and of models vulnerable label ambiguity issues. In this article, we propose self-optimal zero-shot (SOZIL) tackle problems. The contribution SOZIL twofold....