Article
Robotics
Robert DeBortoli, Li Fuxin, Ashish Kapoor, Geoffrey A. Hollinger
Summary: In this work, a 3D adversarial training architecture is proposed to address the problem of 3D object detection from point clouds in data-limited environments. By leveraging an adaptive sampling module to reason about the unstructured nature of point cloud data, the approach encourages the 3D feature encoder to extract features that are invariant across simulated and real scenes, leading to improved object detection performance.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2021)
Article
Robotics
Kenneth Omokhagbo Afebu, Jiyuan Tian, Yang Liu, Evangelos Papatheou, Shyam Prasad
Summary: An AI-assisted dynamic tissue evaluation method is proposed for early diagnosis of bowel cancer by detecting changes in the biomechanical properties of affected lesions. The method involves processing and analyzing dynamic signals from a vibrational capsule in contact with in-situ bowel lesions. Different classification models, including Multi-Layer Perceptron (MLP) and Stacking Ensemble networks (SE), were developed using supervised and unsupervised learning techniques. The results showed that MLPs outperformed SEs in accuracy and number of high-performance models, while unsupervised classification identified two categories representing benign and malignant lesions.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Chemistry, Analytical
Yangqing Ye, Xiaolon Ma, Xuanyi Zhou, Guanjun Bao, Weiwei Wan, Shibo Cai
Summary: A dynamic and real-time object detection algorithm is proposed for home service robots operating indoors. The algorithm consists of an image deblurring algorithm and an object detection algorithm. By improving the clarity of motion-blurred images and employing an attention module and lightweight network unit, the algorithm achieves better detection precision and computational efficiency for small and occluded objects.
Article
Computer Science, Artificial Intelligence
Siyuan Ren, Xiao Pan, Wenjie Zhao, Binling Nie, Bo Han
Summary: In this paper, a dynamic graph transformer 3D object detection network (DGT-Det3D) based on a dynamic graph transformer (DGT) module and a proposal-aware fusion (PAF) module is proposed to address the challenges of point-based 3D object detection. Experimental results demonstrate that the proposed method achieves state-of-the-art accuracy on the KITTI dataset and brings significant improvements when combined with other methods.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Robotics
Issa Mouawad, Nikolas Brasch, Fabian Manhardt, Federico Tombari, Francesca Odone
Summary: Monocular 3D object detection is cost-effective and widely available, but annotation complexity limits dataset size. This research proposes a self-supervised method that uses temporal consistency of object poses as a supervision signal to improve detection performance by refining pose predictions and generating high-quality pseudo labels.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Mohammed Ayoub Alaoui Mhamdi, Djemel Ziou
Summary: This paper presents a critical points based descriptor for 3 D objects recognition, utilizing a size function to represent critical points and the links between them, and a metric learning method to deal with partial matching problems. Experimental results show that the proposed method performs well in 3 D object recognition.
PATTERN RECOGNITION
(2021)
Article
Robotics
Zehan Zhang, Yang Ji, Wei Cui, Yulong Wang, Hao Li, Xian Zhao, Duo Li, Sanli Tang, Ming Yang, Wenming Tan, Shiliang Pu
Summary: This article presents a semi-supervised 3D object detection framework for outdoor scenes, proposing an adaptive thresholds search method and an iterative training mechanism to improve performance. Experimental results demonstrate the outstanding performance of this framework in benchmark testing.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Multidisciplinary Sciences
Shenming Qu, Xinyu Yang, Yiming Gao, Shengbin Liang
Summary: 3D object detection is crucial for autonomous driving perception. Current monocular 3D object detection methods use RGB images and pseudo radar point clouds as input, but they have limitations such as not being able to utilize depth and semantic information simultaneously. We propose a dynamic convolution guided by depth map approach to improve monocular 3D object detection, which significantly enhances detection accuracy.
Article
Chemistry, Multidisciplinary
Yanxin Hu, Gang Liu, Zhiyu Chen, Jianwei Guo
Summary: This paper proposes an object detection algorithm based on a combination of improved YOLOv4 and improved GhostNet, which can balance detection accuracy and efficiency on resource-limited devices and has better performance compared to traditional methods.
APPLIED SCIENCES-BASEL
(2022)
Article
Robotics
Peixuan Li, Huaici Zhao
Summary: In this work, the novel one-stage and keypoint-based framework KM3D-Net is proposed for monocular 3D object detection using only RGB images, achieving superior efficiency and accuracy on the popular KITTI dataset. Additionally, this is the first successful application of semi-supervised learning in monocular 3D object detection, surpassing most previous fully supervised methods with minimal labeled data.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2021)
Article
Robotics
Hyungseok Kim, Hyeongjin Kim, Seonil Lee, Hyeonbeom Lee
Summary: This article proposes a new exploration algorithm based on the path traveled by a mobile robot and segmenting a two-dimensional map, as well as an object detection-based exploration algorithm to reduce the probability of collision with 3D obstacles. Simulation and experiment results demonstrate that the algorithm can significantly shorten the exploration path and time.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Automation & Control Systems
Shaopeng Liu, Guohui Tian, Ying Zhang, Mengyang Zhang, Shuo Liu
Summary: This article focuses on active object detection (AOD) and proposes a novel deep-Q-learning-network-based approach to enhance the training efficiency and testing accuracy. A new reward function and a long-term learning strategy are also introduced to improve performance and adaptability. Experimental results demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2022)
Article
Robotics
Zhongxiang Zhou, Yifei Yang, Yue Wang, Rong Xiong
Summary: Detecting both known and unknown objects is essential for robots in unstructured environments. In this letter, we propose Openset RCNN, a method to address open-set object detection challenges. We use a classification-free region proposal network (CF-RPN) to separate unknown objects and background, and a prototype learning network (PLN) to represent unknown objects in a latent space. We introduce a new benchmark dataset for evaluation and demonstrate the effectiveness of our approach through extensive experiments. Our Openset RCNN enables robots to perceive open-set objects in cluttered environments.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Robotics
Yao Li, Jiajun Deng, Yu Zhang, Jianmin Ji, Houqiang Li, Yanyong Zhang
Summary: A recent trend is to combine multiple sensors (i.e., cameras, LiDARs and millimeter-wave Radars) to achieve robust multi-modal perception for autonomous systems such as self-driving vehicles. However, a systematic study on how to integrate these three types of sensors to develop effective multi-modal 3D object detection and tracking is still missing. This research carefully analyzes the strengths and weaknesses of each data modality, compares different fusion strategies, and proposes a simple yet effective multi-modal 3D object detection and tracking framework. Extensive experiments on the nuScenes dataset demonstrate that the proposed framework achieves remarkable improvements over the LiDAR-only baseline and comparable performance with state-of-the-art fusion-based methods.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Robotics
Lin Zhao, Meiling Wang, Yufeng Yue
Summary: Camera-LiDAR fusion is a promising option for 3D vehicle detection in autonomous driving scenarios. Traditional camera-LiDAR based methods may perform poorly when lacking semantic segmentation labels. To address this issue, we propose a novel semantic augmentation method that improves the performance of multimodal 3D vehicle detection and achieves significant improvements in challenging detection scenarios.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Mathematics
Pamela Chinas-Sanchez, Ismael Lopez-Juarez, Jose Antonio Vazquez-Lopez, Abdelkader El Kamel, Jose Luis Navarro-Gonzalez
Summary: This paper presents a methodology using Mahalanobis distance and Support Vector Machine to recognize two multivariate patterns. By considering the correlation and stability of variables, recognition rates of up to 91.6% were achieved.
Article
Computer Science, Information Systems
Victor Lomas-Barrie, Mario Pena-Cabrera, Ismael Lopez-Juarez, Jose Luis Navarro-Gonzalez
Summary: This article presents the design and implementation of an invariant object recognition system, which utilizes a neural network to compute the Boundary Object Function for object classification. By leveraging parallel computing, the system is able to rapidly encode, describe, and predict objects during video capture, leading to reduced latency.
Article
Mathematics
Pamela Chinas-Sanchez, Ismael Lopez-Juarez, Jose Antonio Vazquez-Lopez, Jose Luis Navarro-Gonzalez, Aidee Hernandez-Lopez
Summary: This paper presents an innovative method based on MVPR and process monitoring to predict possible faults by identifying distinctive out-of-control multivariate patterns in an automated process. Results from the application in robotic GMAW show an overall accuracy of up to 88.8%, demonstrating the effectiveness of the method.
Article
Mathematics
Jose Ruiz-Tamayo, Jose Antonio Vazquez-Lopez, Edgar Augusto Ruelas-Santoyo, Aidee Hernandez-Lopez, Ismael Lopez-Juarez, Armando Javier Rios-Lira
Summary: The proposed model in this research overcomes the limitations of traditional Multivariate Statistical Process Control (MSPC) by using techniques like dimensional reduction and Bayesian Inference to identify the relationship between multivariate and univariate variables for process monitoring. The model's efficiency was found to be superior to the Hotelling's T-2 chart method in experimental results, validating its effectiveness.
Article
Chemistry, Multidisciplinary
David Ortega-Aranda, Julio Fernando Jimenez-Vielma, Baidya Nath Saha, Ismael Lopez-Juarez
Summary: Research has shown that for robot assembly applications in industry, using dual arm robot systems to design and explore applications based on flexibility is beneficial. However, reducing the use of fixtures increases uncertainty in component location and the potential for collisions during the assembly process. In response to this, the addition of perception devices such as force/torque sensors is being studied to generate useful data for control actions.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Interdisciplinary Applications
Alan Maldonado-Ramirez, Reyes Rios-Cabrera, Ismael Lopez-Juarez
Summary: Manufacturing companies need to enhance agility to meet market demands, with the flexibility and vision-based perception technology of robots enabling the creation of adaptive industrial robots. The use of deep reinforcement learning methods can solve path-following tasks and achieve smooth operations on industrial robots.
ROBOTICS AND COMPUTER-INTEGRATED MANUFACTURING
(2021)
Article
Computer Science, Information Systems
G. M. Mendez, Ismael Lopez-Juarez, P. N. Montes-Dorantes, M. A. Garcia
Summary: This paper presents a new method for constructing and training Interval Type-3 Fuzzy Logic Systems with uncertain inputs, called IT3 NSFLS-2. The methodology includes two processes: constructing the structure of the IT3 NSFLS-2 systems based on the level-alpha-0 of the IT2 FLS and the secondary membership function using Gaussian modeling, and training the parameters of the alpha-0 FRB using the gradient descent algorithm. The proposed method was applied to predict the surface temperature of a transfer bar in an industrial hot strip mill facility, and it outperformed similar methods in terms of stability and performance.
Proceedings Paper
Engineering, Multidisciplinary
Roman Osorio-Comparan, Diego A. Nava, Ismael Lopez-Juarez, Victor Lomas, Hector Kaschel, Cristian Ahumada, Gaston Lefranc
Summary: This article aims to establish a computer vision system for object recognition in a given plan, enabling a KUKA KR-5 industrial robot to manipulate objects. Three algorithms are evaluated for performance improvement, with the use of a Microsoft Kinect 2.0 sensor as input device. Object recognition is achieved through FastOrb and Brisk algorithms, along with Ransac-PCA algorithm for object detection in a three-dimensional working environment. The Moveit algorithm is used to obtain trajectories and calculate inverse kinematics, while Gazebo is utilized for simulating the KUKA KR-5 industrial manipulator.
2021 IEEE CHILEAN CONFERENCE ON ELECTRICAL, ELECTRONICS ENGINEERING, INFORMATION AND COMMUNICATION TECHNOLOGIES (IEEE CHILECON 2021)
(2021)
Article
Computer Science, Information Systems
Angel Rodriguez-Linan, Ismael Lopez-Juarez, Alan Maldonado-Ramirez, Antonio Zalapa-Elias, Luis Torres-Trevino, Jose Luis Navarro-Gonzalez, Pamela Chinas-Sanchez
Summary: This paper presents an approach for robots to acquire path-following skills through human interaction, focusing on mapping links and joints from a human operator to the robot. The method is validated using a motion capture system to evaluate spatial deviation from the human-taught path to the robot's final trajectory.
Article
Computer Science, Information Systems
Qing Shi, Jin Zhao, Abdelkader El Kamel, Ismael Lopez-Juarez
Summary: This paper establishes a hierarchical architecture for trajectory planning and control for safe driving with multiple participants using time-varying model predictive control methodology. A high-level planner formulates an optimal control problem to obtain an optimal trajectory, while a low-level controller computes an appropriate steering angle to execute the planned maneuver. This framework combines different constraints in each optimal control problem, ensuring feasibility of safe trajectory planning and tracking stability for safe driving.
Article
Engineering, Multidisciplinary
Arturo Juarez-Hernandez, Gerardo Trapaga-Martinez, Jose Luis Camacho-Martinez, Carlos Gonzalez-Rivera, Ismael Lopez Juarez
INDIAN JOURNAL OF ENGINEERING AND MATERIALS SCIENCES
(2020)
Article
Engineering, Multidisciplinary
Aidee Hernandez-Lopez, Jose-Antonio Vazquez-Lopez, Ismael Lopez-Juarez, Roberto Baeza-Serrato, Jose-Amir Gonzalez-Calderon
Proceedings Paper
Computer Science, Information Systems
J. Carlos Mariscal-Gomez, Roman Osorio-Comparan, Cesar Serrano, Kaori Becerril, Ismael Lopez-Juarez, Miguel Bustamante, Gaston Lefranc
2019 IEEE CHILEAN CONFERENCE ON ELECTRICAL, ELECTRONICS ENGINEERING, INFORMATION AND COMMUNICATION TECHNOLOGIES (CHILECON)
(2019)
Proceedings Paper
Computer Science, Information Systems
M. Bustamante, A. Rienzo, R. Osorio-Comparan, I. Lopez-Juarez, M. Pena, F. Araya, G. Lefranc
2019 IEEE CHILEAN CONFERENCE ON ELECTRICAL, ELECTRONICS ENGINEERING, INFORMATION AND COMMUNICATION TECHNOLOGIES (CHILECON)
(2019)