A5010637524- Prosthetics and Rehabilitation Robotics
- Soft Robotics and Applications
- 3D Shape Modeling and Analysis
- Muscle activation and electromyography studies
- Stroke Rehabilitation and Recovery
- Computer Graphics and Visualization Techniques
- Advanced Vision and Imaging
- Robot Manipulation and Learning
- Surgical Simulation and Training
- Anomaly Detection Techniques and Applications
- Robotics and Sensor-Based Localization
- Image Processing and 3D Reconstruction
- Advanced MRI Techniques and Applications
- Advanced Sensor and Energy Harvesting Materials
- Modular Robots and Swarm Intelligence
- 3D Surveying and Cultural Heritage
- Micro and Nano Robotics
- Mechanical Circulatory Support Devices
- Optical Coherence Tomography Applications
- Integrated Circuits and Semiconductor Failure Analysis
- Adversarial Robustness in Machine Learning
- Vascular Malformations Diagnosis and Treatment
- Reinforcement Learning in Robotics
- Cerebral Palsy and Movement Disorders
- Advanced Materials and Mechanics
North Carolina State University
2021-2025
Northwestern Polytechnical University
2023-2025
Xi'an University of Science and Technology
2024-2025
University of California, San Diego
2017-2024
University of North Carolina at Chapel Hill
2022-2024
City College of New York
2018-2024
City College
2018-2024
Beihang University
2014-2024
Beijing Jingshida Electromechanical Equipment Research Institute
2024
Philips (India)
2024
Point cloud is an important type of geometric data structure. Due to its irregular format, most researchers transform such regular 3D voxel grids or collections images. This, however, renders unnecessarily voluminous and causes issues. In this paper, we design a novel neural network that directly consumes point clouds, which well respects the permutation invariance points in input. Our network, named PointNet, provides unified architecture for applications ranging from object classification,...
Large repositories of 3D shapes provide valuable input for data-driven analysis and modeling tools. They are especially powerful once annotated with semantic information such as salient regions functional parts. We propose a novel active learning method capable enriching massive geometric datasets accurate region annotations. Given shape collection user-specified label our goal is to correctly demarcate the corresponding minimal manual work. Our framework achieves this by cycling between...
Robust low-level image features have been proven to be effective representations for a variety of visual recognition tasks such as object and scene classification; but pixels, or even local patches, carry little semantic meanings. For high level tasks, are potentially not enough. In this paper, we propose high-level representation, called the Object Bank, where an is represented scale-invariant response map large number pre-trained generic detectors, blind testing dataset task. Leveraging on...
We present PartNet: a consistent, large-scale dataset of 3D objects annotated with fine-grained, instance-level, and hierarchical part information. Our consists 573,585 instances over 26,671 models covering 24 object categories. This enables serves as catalyst for many tasks such shape analysis, dynamic scene modeling simulation, affordance others. Using our dataset, we establish three benchmarking evaluating recognition: fine-grained semantic segmentation, instance segmentation. benchmark...
In this paper, we study the problem of semantic annotation on 3D models that are represented as shape graphs. A functional view is taken to represent localized information graphs, so annotations such part segment or keypoint nothing but 0-1 indicator vertex functions. Compared with images 2D grids, graphs irregular and non-isomorphic data structures. To enable prediction functions them by convolutional neural networks, resort spectral CNN method enables weight sharing parametrizing kernels...
We present a learning framework for abstracting complex shapes by to assemble objects using 3D volumetric primitives. In addition generating simple and geometrically interpretable explanations of objects, our also allows us automatically discover exploit consistent structure in the data. demonstrate that method predicting shape representations which can be leveraged obtaining parsing across instances collection constructing an similarity measure. examine applications image-based prediction...
The ability to generate novel, diverse, and realistic 3D shapes along with associated part semantics structure is central many applications requiring high-quality assets or large volumes of training data. A key challenge towards this goal how accommodate diverse shape variations, including both continuous deformations parts as well structural discrete alterations which add to, remove from, modify the constituents compositional structure. Such object can typically be organized into a...
Single image 3D reconstruction is an important but challenging task that requires extensive knowledge of our natural world. Many existing methods solve this problem by optimizing a neural radiance field under the guidance 2D diffusion models suffer from lengthy optimization time, inconsistency results, and poor geometry. In work, we propose novel method takes single any object as input generates full 360-degree textured mesh in feed-forward pass. Given image, first use view-conditioned...
3D object classification using deep neural networks has been extremely successful. As the problem of identifying objects many safety-critical applications, have to be robust against adversarial changes input data set. We present a preliminary evaluation attacks on point cloud classifiers by evaluating that were proposed for 2D images, and extending those reduce perceptibility perturbations in space. also show effectiveness simple defenses attacks. Finally, we attempt explain through...
Stereotaxy is a neurosurgical technique that can take several hours to reach specific target, typically utilizing mechanical frame and guided by preoperative imaging. An error in any one of the numerous steps or deviations target anatomy from plan such as brain shift (up 20 mm), may affect targeting accuracy thus treatment effectiveness. Moreover, because procedure performed through small burr hole opening skull prevents tissue visualization, intervention basically “blind” for operator with...
High-performance actuators are crucial to enable mechanical versatility of wearable robots, which required be lightweight, highly backdrivable, and with high bandwidth. State-of-the-art actuators, e.g., series elastic have compromise bandwidth improve compliance (i.e., backdrivability). In this article, we describe the design human-robot interaction modeling a portable hip exoskeleton based on our custom quasi-direct drive actuation torque density motor low ratio gear). We also present...
This paper presents a fully actuated robotic system for percutaneous prostate therapy under continuously acquired live magnetic resonance imaging (MRI) guidance. The is composed of modular hardware and software to support the surgical workflow intraoperative MRI-guided procedures. We present development 6-degree-of-freedom (DOF) needle placement robot transperineal interventions. consists 3-DOF driver module Cartesian motion module. provides cannula translation rotation (2-DOF) stylet...
This letter presents design principles for comfort-centered wearable robots and their application in a lightweight backdrivable knee exoskeleton. The mitigation of discomfort is treated as mechanical control issues three solutions are proposed this letter: 1) new structure optimizes the strap attachment configuration suit layout to ameliorate excessive shear forces conventional design; 2) rolling joint double-hinge mechanisms reduce misalignment sagittal frontal plane, without increasing...
Soft machines typically exhibit slow locomotion speed and low manipulation strength because of intrinsic limitations soft materials. Here, we present a generic design principle that harnesses mechanical instability for variety spine-inspired fast strong machines. Unlike most current robots are designed as inherently unimodally stable, our leverages tunable snap-through bistability to fully explore the ability rapidly store release energy within tens milliseconds. We demonstrate this with...
Back injuries are the most prevalent work-related musculoskeletal disorders and represent a major cause of disability. Although innovations in wearable robots aim to alleviate this hazard, majority existing exoskeletons obtrusive because rigid linkage design limits natural movement, thus causing ergonomic risk. Moreover, these systems typically only suitable for one type movement assistance, not ubiquitous wide variety activities. To fill gap, letter presents new robot approach continuum...
We propose an adversarial defense method that achieves state-of-the-art performance among attack-agnostic methods while also maintaining robustness to input resolution, scale of perturbation, and dataset size. Based on convolutional sparse coding, we construct a stratified low-dimensional quasi-natural image space faithfully approximates the natural removing perturbations. introduce novel Sparse Transformation Layer (STL) in between first layer neural network efficiently project images into...
In this work, we tackle the problem of category-level online pose tracking objects from point cloud sequences. For first time, propose a unified framework that can handle 9DoF for novel rigid object instances as well per-part articulated known categories. Here pose, comprising 6D and 3D size, is equivalent to amodal bounding box representation with free pose. Given depth at current frame estimated last frame, our end-to-end pipeline learns accurately update Our composed three modules: 1)...
Medical robots can play an important role in mitigating the spread of infectious diseases and delivering quality care to patients during COVID-19 pandemic. Methods procedures involving medical continuum care, ranging from disease prevention, screening, diagnosis, treatment, home have been extensively deployed also present incredible opportunities for future development. This article provides overview current state art, highlighting enabling technologies unmet needs prospective technological...
Recent work [28], [5] has demonstrated that volumetric scene representations combined with differentiable volume rendering can enable photo-realistic for challenging scenes mesh reconstruction fails on. However, these methods entangle geometry and appearance in a "black-box" cannot be edited. Instead, we present an approach explicitly disentangles geometry—represented as continuous 3D volume—from appearance—represented 2D texture map. We achieve this by introducing 3D-to-2D mapping (or...
Magnetic resonance imaging (MRI) can provide high-quality 3-D visualization of target anatomy, surrounding tissue, and instrumentation, but there are significant challenges in harnessing it for effectively guiding interventional procedures. Challenges include the strong static magnetic field, rapidly switching field gradients, high-power radio frequency pulses, sensitivity to electrical noise, constrained space operate within bore scanner. MRI has a number advantages over other medical...
State-of-the-art exoskeletons are typically limited by the low control bandwidth and small-range stiffness of actuators, which based on high gear ratios elastic components (e.g., series actuators). Furthermore, most discrete gait phase detection and/or control, resulting in discontinuous torque profiles. To fill these two gaps, we developed a portable, lightweight knee exoskeleton using quasi-direct-drive (QDD) actuation that provides 14 N·m (36.8% biological joint moment for overground...
Harnessing snapping, an instability phenomenon observed in nature (e.g., Venus flytraps), for autonomy has attracted growing interest autonomous soft robots. However, achieving self-sustained snapping and snapping-driven motions robots remains largely unexplored. Here, harnessing bistable, ribbon ring-like structures realizing a library of liquid-crystal elastomer wavy rings under constant thermal photothermal actuation are reported. The induces continuous ring flipping that drives dancing...
Large, high-capacity models trained on diverse datasets have shown remarkable successes efficiently tackling downstream applications. In domains from NLP to Computer Vision, this has led a consolidation of pretrained models, with general backbones serving as starting point for many Can such happen in robotics? Conventionally, robotic learning methods train separate model every application, robot, and even environment. we instead generalist X-robot policy that can be adapted new robots,...