检索结果-内蒙古大学图书馆

PhysORD: A Neuro-Symbolic Approach for Physics-infused Motion Prediction in Off-road Driving

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhao, Zhipeng Li, Bowen Du, Yi Fu, Taimeng Wang, Chen Lab Institute for Artificial Intelligence and Data Science Department of Computer Science and Engineering University at Buffalo BuffaloNY14260 United States Robotics Institute Carnegie Mellon University PittsburghPA15213 United States

Motion prediction is critical for autonomous off-road driving, however, it presents significantly more challenges than on-road driving because of the complex interaction between the vehicle and the terrain. Traditional physics-based approaches encounter difficulties in accurately modeling dynamic systems and external disturbance. In contrast, data-driven neural networks require extensive datasets and struggle with explicitly capturing the fundamental physical laws, which can easily lead to poor generalization. By merging the advantages of both methods, neuro-symbolic approaches present a promising direction. These methods embed physical laws into neural models, potentially significantly improving generalization capabilities. However, no prior works were evaluated in real-world settings for off-road driving. To bridge this gap, we present PhysORD, a neural-symbolic approach integrating the conservation law, i.e., the Euler-Lagrange equation, into data-driven neural models for motion prediction in off-road driving. Our experiments showed that PhysORD can accurately predict vehicle motion and tolerate external disturbance by modeling uncertainties. The learned dynamics model achieves 46.7% higher accuracy using only 3.1% of the parameters compared to data-driven methods, demonstrating the data efficiency and superior generalization ability of our neural-symbolic method. Copyright © 2024, The Authors. All rights reserved.

关键词： Off road vehicles

Flickr30K-CFQ: A Compact and Fragmented Query Dataset for Text-image Retrieval

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Liu, Haoyu Song, Yaoxian Wang, Xuwu Zhu, Xiangru Li, Zhixu Song, Wei Li, Tiefeng Research Center for Intelligent Robotics Zhejiang Lab China Department of Engineering Mechanics Center for X-Mechanics Zhejiang University China Shanghai Key Laboratory of Data Science School of Computer Science Fudan University China

With the explosive growth of multi-modal information on the Internet, unimodal search cannot satisfy the requirement of Internet applications. Text-image retrieval research is needed to realize high-quality and efficient retrieval between different modalities. Existing text-image retrieval research is mostly based on general vision-language datasets (e.g. MS-COCO, Flickr30K), in which the query utterance is rigid and unnatural (i.e. verbosity and formality). To overcome the shortcoming, we construct a new Compact and Fragmented Query challenge dataset (named Flickr30K-CFQ) to model text-image retrieval task considering multiple query content and style, including compact and fine-grained entity-relation corpus. We propose a novel LLM-based Query-enhanced method using prompt engineering based on LLM. Experiments show that our proposed Flickr30-CFQ reveals the insufficiency of existing vision-language datasets in realistic text-image tasks. Our LLM-based Query-enhanced method applied on different existing text-image retrieval models improves query understanding performance both on public dataset and our challenge set Flickr30-CFQ with over 0.9% and 2.4% respectively. Our project can be available anonymously in https://***/view/Flickr30K-cfq. Copyright © 2024, The Authors. All rights reserved.

关键词： Image enhancement

An Efficient Scene Coordinate Encoding and Relocalization Method

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Xu, Kuan Jiang, Zeyu Cao, Haozhi Yuan, Shenghai Wang, Chen Xie, Lihua School of Electrical and Electronic Engineering Nanyang Technological University 50 Nanyang Avenue 639798 Singapore Spatial AI & Robotics Lab Department of Computer Science and Engineering University at Buffalo BuffaloNY14260 United States

Scene Coordinate Regression (SCR) is a visual localization technique that utilizes deep neural networks (DNN) to directly regress 2D-3D correspondences for camera pose estimation. However, current SCR methods often face challenges in handling repetitive textures and meaningless areas due to their reliance on implicit triangulation. In this paper, we propose an efficient scene coordinate encoding and relocalization method. Compared with the existing SCR methods, we design a unified architecture for both scene encoding and salient keypoint detection, enabling our system to focus on encoding informative regions, thereby significantly enhancing efficiency. Additionally, we introduce a mechanism that leverages sequential information during both map encoding and relocalization, which strengthens implicit triangulation, particularly in repetitive texture environments. Comprehensive experiments conducted across indoor and outdoor datasets demonstrate that the proposed system outperforms other state-of-the-art (SOTA) SCR methods. Our single-frame relocalization mode improves the recall rate of our baseline by 6.4% and increases the running speed from 56Hz to 90Hz. Furthermore, our sequence-based mode increases the recall rate by 11% while maintaining the original efficiency. Copyright © 2024, The Authors. All rights reserved.

关键词： Triangulation

A Hybrid EEG-EMG Framework for Humanoid Control using Deep Learning Transformers

学校读者我要写书评

暂无评论

A Hybrid EEG-EMG Framework for Humanoid Control using Deep L...

IEEE International Conference on robotics and Biomimetics

作者： Parthan Olikkal Branesh M. Pillai Jackrit Suthakorn Habib Ali Ramana Vinjamuri Department of Computer Science and Electrical Engineering Vinjamuri lab University of Maryland Baltimore County Baltimore MD USA Center for Biomedical and Robotics Technology Mahidol University Salaya Nakorm Pathom Thailand

ISBN: (数字)9798331509644

ISBN: (纸本)9798331509651

Brain-computer interfaces (BCIs) offer promising solutions for upper limb rehabilitation. Despite advancements in deep learning, traditional models for motor rehabilitation using electroencephalography (EEG) or electromyography (EMG) to control assistive devices require enhancement. This study aims to enhance motor control capabilities by integrating EEG and EMG signals using a Transformer-based deep learning model. Ten able-bodies subjects performed center-out tasks on a low-cost upper limb rehabilitation table, capturing 2D kinematic data, EEG, and EMG signals simultaneously. The tasks varied in complexity across four levels. Preprocessed EEG and EMG signals were fused and given as input to the proposed model, which was evaluated using three performance metrics. Results showed that the EEG-EMG combined model achieved 87.27% accuracy across all the four levels. Furthermore, the model's output successfully controlled a humanoid robot to replicate similar movements. These findings highlight the efficacy of combined EEG-EMG data in improving accuracy and performance in BCI applications, advancing assistive technologies and neurorehabilitation interventions.

关键词： Deep learning Accuracy Humanoid robots Brain modeling Transformers Motors Electromyography Electroencephalography Data models Brain-computer interfaces

Long-Term Invariant Local Features via Implicit Cross-Domain Correspondences

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Pataki, Zador Altillawi, Mohammad Kanakis, Menelaos Pautrat, Rémi Shen, Fengyi Liu, Ziyuan Van Gool, Luc Pollefeys, Marc The Computer Vision and Geometry Lab Department of Computer Science ETH Zurich Switzerland The Computer Vision Center CVC-Barcelona The Intelligent Robotics Cloud Technology lab of Huawei-Munich Germany The Computer Vision Lab Department Electrical Engineering ETH Zurich Switzerland The Intelligent Robotics Cloud Technology lab of Huawei-Munich Germany The Intelligent Robotics Cloud Technology lab of Huawei-Munich Germany The Center for Processing Speech and Images KU Leuven The Computer Vision Lab ETH Zurich Switzerland

Modern learning-based visual feature extraction networks perform well in intra-domain localization, however, their performance significantly declines when image pairs are captured across long-term visual domain variations, such as different seasonal and daytime variations. In this paper, our first contribution is a benchmark to investigate the performance impact of long-term variations on visual localization. We conduct a thorough analysis of the performance of current state-of-the-art feature extraction networks under various domain changes and find a significant performance gap between intra- and cross-domain localization. We investigate different methods to close this gap by improving the supervision of modern feature extractor networks. We propose a novel data-centric method, Implicit Cross-Domain Correspondences (iCDC). iCDC represents the same environment with multiple Neural Radiance Fields, each fitting the scene under individual visual domains. It utilizes the underlying 3D representations to generate accurate correspondences across different long-term visual conditions. Our proposed method enhances cross-domain localization performance, significantly reducing the performance gap. When evaluated on popular long-term localization benchmarks, our trained networks consistently outperform existing methods. This work serves as a substantial stride toward more robust visual localization pipelines for long-term deployments, and opens up research avenues in the development of long-term invariant descriptors. Copyright © 2023, The Authors. All rights reserved.

关键词： Feature extraction

FDCT: Fast Depth Completion for Transparent Objects

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Li, Tianan Chen, Zhehan Liu, Huan Wang, Chen The School of Mechanical Engineering University of Science and Technology Beijing Beijing100083 China The Department of Electrical and Computer Engineering McMaster University Canada The Spatial AI & Robotics Lab The Department of Computer Science and Engineering State University of New York BuffaloNY14260 United States

Depth completion is crucial for many robotic tasks such as autonomous driving, 3-D reconstruction, and manipulation. Despite the significant progress, existing methods remain computationally intensive and often fail to meet the real-time requirements of low-power robotic platforms. Additionally, most methods are designed for opaque objects and struggle with transparent objects due to the special properties of reflection and refraction. To address these challenges, we propose a Fast Depth Completion framework for Transparent objects (FDCT), which also benefits downstream tasks like object pose estimation. To leverage local information and avoid overfitting issues when integrating it with global information, we design a new fusion branch and shortcuts to exploit low-level features and a loss function to suppress overfitting. This results in an accurate and user-friendly depth rectification framework which can recover dense depth estimation from RGB-D images alone. Extensive experiments demonstrate that FDCT can run about 70 FPS with a higher accuracy than the state-of-the-art methods. We also demonstrate that FDCT can improve pose estimation in object grasping tasks. The source code is available at https://***/Nonmy/FDCT. Copyright © 2023, The Authors. All rights reserved.

关键词： Deep learning

Robots in education: Influence of Regulatory Focus Theory

学校读者我要写书评

暂无评论

Robots in education: Influence of Regulatory Focus Theory

IEEE International Workshop on Robot and Human Communication (ROMAN)

作者： Xiaoxuan Hei Heng Zhang Adriana Tapus Computer Science and System Engineering Department (U2IS) Autonomous Systems and Robotics Lab ENSTA Paris Institut Polytechnique de Paris Paris France

The Covid-19 pandemic has massively developed the use of distance learning. The limits of this practice have gradually come to light, both for students and for teachers. It is now crucial to design alternative solutions to overcome the shortcomings of videoconferencing in terms of involvement, concentration, learning, and equity. Social robots are increasingly used as tutors in the educational context and help improve teaching efficiency. Many psychology-based principles have been applied in education to guide instructional strategies, motivate students, and create a positive and productive learning environment. In this work, we use Regulatory Focus Theory (RFT), which categorizes an individual’s motivation into two types: Promotion and Prevention. Promotion-focused individuals are motivated by the potential for growth and achievement, whereas prevention-focused individuals are motivated by the potential for avoiding negative outcomes. Based on RFT, we aim to explore if and how the regulatory-focused behavior of the tutor robot can affect participants’ learning outcomes. In this work, a language learning scenario was designed with two conditions: (1) a robot tutor with promotion-focused behavior, (2) a robot tutor with prevention-focused behavior. The results are encouraging and support that promotion robot tutor can increase the learning efficiency of promotion participants and prevention robot tutor will enhance the learning interest of prevention participants.

关键词：

Learning Inverse Kinematics Multiplicity of Concentric Tube Robots Using Invertible Neural Networks

学校读者我要写书评

暂无评论

Learning Inverse Kinematics Multiplicity of Concentric Tube ...

International Symposium on Medical robotics (ISMR)

作者： Paul H. Kang Radian Gondokaryono Majid Roshanfar Robert H. Nguyen Thomas Looi James M. Drake Dale Podolsky Institute of Biomedical Engineering University of Toronto Canada Wilfred and Joyce Posluns Centre for Image Guided Innovation and Therapeutic Intervention Hospital for Sick Children Toronto Canada Department of Computer Science Medical Computer Vision and Robotics Lab University of Toronto Canada

ISBN: (数字)9798331599003

ISBN: (纸本)9798331599010

Concentric tube robots (CTR) are a promising technology for medical applications due to their small size, flexibility, and ability to make complex shapes. These robots are built from a series of pre-curved, super-elastic tubes that are arranged concentrically and manipulated through rotational and translational movements at their proximal ends. Achieving accurate kinematics is essential in making CTRs useful in minimally invasive surgical procedures where precision and safety is paramount. Due to the difficulty of incorporating nonlinear effects like friction and tube clearances into analytical models, previous works have investigated machine learningbased models for CTR kinematics, leading to higher kinematic accuracies. We present a kinematic model for CTRs using an invertible neural network architecture that, unlike other learningbased models, can generate multiple inverse kinematic solutions. Our model achieved a mean forward kinematic tip error of 2.86 mm ( $\mathbf{3. 4 3 \%}$ normalized to arclength), outperforming a Cosserat rod-based analytical CTR model. In three test cases (two static points and a circle trajectory), our model achieved mean inverse kinematic errors of $6.19,3.33$ , and $3.86 ~\text{mm}(5.66 \%, 4.12 \%$ , and 4.16 % normalized to arclength). We additionally present a robust data capture pipeline that is able to reconstruct the CTR's shape that uses state-of-the-art segmentation models.

关键词： Analytical models Concentric tube robots Accuracy Minimally invasive surgery Shape Neural networks Pipelines Kinematics Trajectory Safety

Towards the Influence of Human Observer Eye-Movements on Discriminating Between Perceived Sociability of Different Robots 54

学校读者我要写书评

暂无评论

Towards the Influence of Human Observer Eye-Movements on Dis...

54th International Symposium on robotics, ISR Europe 2022

作者： Kühnlenz, Kolja Kühnlenz, Barbara Robotics Research Lab Department of Electrical Engineering and Computer Science Coburg University of Applied Sciences and Arts CoburgD-96450 Germany Academic Center for Sciences and Humanities Coburg University of Applied Sciences and Arts CoburgD-96450 Germany

ISBN: (纸本)9783800758920

Results from a study on Perceived Sociability of a robot dependent on different gaze behaviors of human observers are presented. A 2x2 between-subjects design is used with independent variables’gaze-behavior’ (static/dynamic) and’robot’ (appearance) of a displayed robot head. Participants are told to fixate a point on a screen, which is either static or moving linearly, while a picture of one of two robot heads is shown in the covert visual attention region of the human observer. The robots differ in the level of human-likeness (mechanistic vs. anthropomorphic). Significant results show, that Perceived Sociability depends on the interaction of gaze-behavior and robot appearance. The findings show, that eye-movements influence covert perception of robot attributes and contribute to a better understanding of gaze dependent perception of robot appearance with potential practical implications of situational perception of anthropomorphic attributes like sociability situationally impacting interaction with the robot. © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach.

关键词： Eye movements