检索结果-内蒙古大学图书馆

Fast and Continuous Foothold Adaptation for Dynamic Locomotion Through CNNs

IEEE robotics AND automation LETTERS 2019年第2期4卷 2140-2147页

作者： Magana, Octavio Antonio Villarreal Barasuol, Victor Camurri, Marco Franceschi, Luca Focchi, Michele Pontil, Massimiliano Caldwell, Darwin G. Semini, Claudio Ist Italiano Tecnol Dynam Legged Syst Lab I-16163 Genoa Italy Univ Oxford Oxford Robot Inst Oxford OX2 6NN England Ist Italian Tecnol Computat Stat & Machine Learning I-16163 Genoa Italy Ist Italiano Tecnol Dept Adv Robot I-16163 Genoa Italy

Legged robots can outperform wheeled machines for most navigation tasks across unknown and rough terrains. For such tasks, visual feedback is a fundamental asset to provide robots with terrain awareness. However, robust dynamic locomotion on difficult terrains with real-time performance guarantees remains a challenge. We present here a real-time, dynamic foothold adaptation strategy based on visual feedback. Our method adjusts the landing position of the feet in a fully reactive manner, using only on-board computers and sensors. The correction is computed and executed continuously along the swing phase trajectory of each leg. To efficiently adapt the landing position, we implement a self-supervised foothold classifier based on a convolutional neural network. Our method results in an up to 200 times faster computation with respect to the full-blown heuristics. Our goal is to react to visual stimuli from the environment, bridging the gap between blind reactive locomotion and purely vision-based planning strategies. We assess the performance of our method on the dynamic quadruped robot HyQ, executing static and dynamic gaits (at speeds up to 0.5 m/s) in both simulated and real scenarios;the benefit of safe foothold adaptation is clearly demonstrated by the overall robot behavior.

关键词： Legged Robots Reactive and Sensor-Based Planning deep learning in robotics and automation

来源：评论

学校读者我要写书评

暂无评论

learning Long-Range Perception Using Self-Supervision From Short-Range Sensors and Odometry

引用

IEEE robotics AND automation LETTERS 2019年第2期4卷 1279-1286页

作者： Nava, Mirko Guzzi, Jerome Chavez-Garcia, R. Omar Gambardella, Luca M. Giusti, Alessandro USI SUPSI Dalle Molle Inst Artificial Intelligence IDSIA CH-6928 Lugano Switzerland

We introduce a general self-supervised approach to predict the future outputs of a short-range sensor (such as a proximity sensor) given the current outputs of a long-range sensor (such as a camera). We assume that the former is directly related to some piece of information to be perceived (such as the presence of an obstacle in a given position), whereas the latter is information rich but hard to interpret directly. We instantiate and implement the approach on a small mobile robot to detect obstacles at various distances using the video stream of the robot's forward-pointing camera, by training a convolutional neural network on automatically-acquired datasets. We quantitatively evaluate the quality of the predictions on unseen scenarios, qualitatively evaluate robustness to different operating conditions, and demonstrate usage as the sole input of an obstacle-avoidance controller. We additionally instantiate the approach on a different simulated scenario with complementary characteristics, to exemplify the generality of our contribution.

关键词： Range sensing computer vision for other robotic applications deep learning in robotics and automation

来源：评论

学校读者我要写书评

暂无评论

Real-Time 3-D Shape Instantiation for Partially Deployed Stent Segments From a Single 2-D Fluoroscopic Image in Fenestrated Endovascular Aortic Repair

引用

IEEE robotics AND automation LETTERS 2019年第4期4卷 3703-3710页

作者： Zheng, Jian-Qing Zhou, Xiao-Yun Riga, Celia Yang, Guang-Zhong Imperial Coll London Hamlyn Ctr Robot Surg London SW7 2AZ England St Marys Hosp Reg Vasc Unit London W2 1NY England Imperial Coll London Acad Div Surg London SW7 2A2 England

In fenestrated endovascular aortic repair (FEVAR), accurate alignment of stent graft fenestrations or scallops with aortic branches is essential for establishing complete blood flow perfusion. Current navigation is largely based on two-dimensional (2-D) fluoroscopic images, which lacks 3-D anatomical information, thus causing a longer operation time and high risks of radiation exposure. Previously, 3-D shape instantiation frameworks for realtime 3-D shape reconstruction of fully deployed or fully compressed stent grafts from a single 2-D fluoroscopic image have been proposed for 3-D navigation in FEVAR. However, thesemethods could not instantiate partially deployed stent segments, as the 3-D marker references are unknown. In this letter, an adapted graph convolutional network (GCN) is proposed to predict 3-D partially deployed marker references from 3-D fully deployed marker references. As the original GCN is for classification, in this letter, the coarsening layers are removed and the softmax function at the network end is replaced with linear mapping for regression. The derived 3-D marker references and the 2-D marker positions are used to instantiate the partially deployed stent segment, combined with the previous 3-D shape instantiation framework. Validations were performed on three typical stent grafts and five patient-specific 3-D printed aortic aneurysm phantoms. Reasonable performances with average mesh distance errors from 1.0 to 2.4 mm and average angular errors around 7.2. were achieved.

关键词： deep learning in robotics and automation computer vision for medical robotics surgical robotics: planning visual-based navigation motion and path planning

来源：评论

学校读者我要写书评

暂无评论

A deep learning Approach for Probabilistic Security in Multi-Robot Teams

引用

IEEE robotics AND automation LETTERS 2019年第4期4卷 4262-4269页

作者： Wehbe, Remy Williams, Ryan K. Virginia Polytech Inst & State Univ Dept Elect & Comp Engn Blacksburg VA 24061 USA

In this letter, we train a convolutional neural network (CNN) to predict the probability of security of a multi-robot system (MRS) when robot interactions are probabilistic. In the context of MRSs, probabilistic security is defined using the control-theoretic notion of left invertibility, a necessary and sufficient condition to avoid perfect attacks. As the probabilistic security problem is NP-Complete, current solutions fail to generalize as the size of the MRS increases. Fortunately, deep neural networks have shown promising results in the efficient computation of solutions to hard problems, which motivates our CNN-based approach. In this context, formulating a method for data generation is non-trivial due to the large space of available interaction graph topologies and training biases introduced by random sampling. As such, we use a two-step approach for data generation where we first explore the space of available topologies, then populate the sampled topologies with probability distributions, all while preventing any biases from occurring in the data. We then train a CNN with convolution layers specifically tailored for graph adjacency matrices. Finally, the validity of our results is demonstrated through simulations.

关键词： Multi-Robot Systems deep learning in robotics and automation Distributed Robot Systems Networked Robots

来源：评论

学校读者我要写书评

暂无评论

learning Affordance Segmentation for Real-World Robotic Manipulation via Synthetic Images

引用

IEEE robotics AND automation LETTERS 2019年第2期4卷 1140-1147页

作者： Chu, Fu-Jen Xu, Ruinian Vela, Patricio A. Georgia Inst Technol Inst Robot & Intelligent Machines Atlanta GA 30332 USA

This letter presents a deep learning framework to predict the affordances of object parts for robotic manipulation. The framework segments affordance maps by jointly detecting and localizing candidate regions within an image. Rather than requiring annotated real-world images, the framework learns from synthetic data and adapts to real-world data without supervision. Themethod learns domain-invariant region proposal networks and task-level domain adaptation components with regularization on the predicted domains. A synthetic version of the UMD data set is collected for autogenerating annotated, synthetic input data. Experimental results show that the proposed method outperforms an unsupervised baseline, and achieves performance close to state-of-the-art supervised approaches. An ablation study establishes the performance gap between the proposedmethod and the supervised equivalent (30%). Real-world manipulation experiments demonstrate use of the affordance segmentations for task execution, which achieves the same performance with supervised approaches.

关键词： Perception for Grasping and Manipulation deep learning in robotics and automation RGB-D Perception

来源：评论

学校读者我要写书评

暂无评论

Vision-Based Estimation of Driving Energy for Planetary Rovers Using deep learning and Terramechanics

引用

IEEE robotics AND automation LETTERS 2019年第4期4卷 3876-3883页

作者： Higa, Shoya Iwashita, Yumi Otsu, Kyohei Ono, Masahiro Lamarre, Olivier Didier, Annie Hoffmann, Mark CALTECH Jet Prop Lab 4800 Oak Grove Dr Pasadena CA 91109 USA Univ Toronto Inst Aerosp Studies STARS Lab Toronto ON M3H 5T6 Canada

This letter presents a prediction algorithm of driving energy for future Mars rover missions. The majority of future Mars rovers would be solar-powered, which would require energy-optimal driving to maximize the range with limited energy. The essential and arguably the most challenging technology for realizing energy-optimal driving is the capability to predict the driving energy, which is needed to construct an energy-aware cost function for path planning. In this letter, we propose vision-based algorithms to remotely predict the driving energy consumption using machine learning. Specifically, we develop and compare two machine-learning models in this letter, namely VeeGer-Energy Net and Veeger-TerramechanicsNet, respectively. The former is trained directly using recorded power, while the latter estimates terrain parameters from the images using a simplified-terramechanics model, and calculate the power based on the model. The two approaches are fully automated self-supervised learning algorithms. To combine RGB and depth images efficiently with high accuracy, we propose a new network architecture called Two-PNASNet-5, which is based on PNASNet-5. We collected a new dataset to verify the effectiveness of the proposed approaches. Comparison of the two approaches showed that Veeger-TerramechanicsNet had better performance than VeeGer-EnergyNet.

关键词： Space robotics and automation wheeled robotics deep learning in robotics and automation

来源：评论

学校读者我要写书评

暂无评论

learning to See the Wood for the Trees: deep Laser Localization in Urban and Natural Environments on a CPU

引用

IEEE robotics AND automation LETTERS 2019年第2期4卷 1327-1334页

作者： Tinchev, Georgi Penate-Sanchez, Adrian Fallon, Maurice Univ Oxford Oxford Robot Inst Dynam Syst Grp Oxford OX2 6NN England

Localization in challenging, natural environments, such as forests or woodlands, is an important capability for many applications from guiding a robot navigating along a forest trail to monitoring vegetation growth with handheld sensors. In this letter, we explore laser-based localization in both urban and natural environments, which is suitable for online applications. We propose a deep learning approach capable of learning meaningful descriptors directly from three-dimensional point clouds by comparing triplets (anchor, positive, and negative examples). The approach learns a feature space representation for a set of segmented point clouds that are matched between a current and previous observations. Our learning method is tailored toward loop closure detection resulting in a small model that can be deployed using only a CPU. The proposed learning method would allow the full pipeline to run on robots with limited computational payloads, such as drones, quadrupeds, or Unmanned Ground Vehicles (UGVs).

关键词： Localization deep learning in robotics and automation visual learning SLAM field robots

来源：评论

学校读者我要写书评

暂无评论

Fully Automated Annotation With Noise-Masked Visual Markers for deep-learning-Based Object Detection

引用

IEEE robotics AND automation LETTERS 2019年第2期4卷 1972-1977页

作者： Kiyokawa, Takuya Tomochika, Keita Takamatsu, Jun Ogasawara, Tsukasa Nara Inst Sci & Technol Div Informat Sci Nara 6300192 Japan

Automated factories use deep-learning-based vision systems to accurately detect various products. However, training such vision systems requires manual annotation of a significant amount of data to optimize the large number of parameters of the deep convolutional neural networks. Such manual annotation is very time-consuming and laborious. To reduce this burden, we propose a fully automated annotation approach without any manual intervention. To do this, we associate one visual marker with one object and capture them in the same image. However, if an image showing the marker is used for training, normally, the neural network learns the marker as a feature of the object. By hiding the marker with a noise mask, we succeeded in reducing this erroneous learning. Experiments verified the effectiveness of the proposed method in comparison with manual annotation, in terms of both the time needed to collect training data and the resulting detection accuracy of the vision system. The time required for data collection was reduced from 16.1 to 1.87 h. The accuracy of the vision system trained with the proposed method was 87.3%, which is higher than the accuracy of a vision system trained with the manual method.

关键词： Computer vision for automation deep learning in robotics and automation object detection segmentation and categorization

来源：评论

学校读者我要写书评

暂无评论

PRIMAL: Pathfinding via Reinforcement and Imitation Multi-Agent learning

引用

IEEE robotics AND automation LETTERS 2019年第3期4卷 2378-2385页

作者： Sartoretti, Guillaume Kerr, Justin Shi, YunFei Wagner, Glenn Kumar, T. K. Satish Koenig, Sven Choset, Howie Carnegie Mellon Univ Robot Inst Pittsburgh PA 15213 USA CSIRO Pullenvale Qld 4069 Australia Univ Southern Calif Comp Sci Dept Los Angeles CA 90089 USA

Multi-agent path finding (MAPF) is an essential component of many large-scale, real-world robot deployments, from aerial swarms to warehouse automation. However, despite the community's continued efforts, most state-of-the-art MAPF planners still rely on centralized planning and scale poorly past a few hundred agents. Such planning approaches are maladapted to real-world deployments, where noise and uncertainty often require paths be recomputed online, which is impossible when planning times are in seconds to minutes. We present PRIMAL, a novel framework for MAPF that combines reinforcement and imitation learning to teach fully decentralized policies, where agents reactively plan paths online in a partially observable world while exhibiting implicit coordination. This framework extends our previous work on distributed learning of collaborative policies by introducing demonstrations of an expert MAPF planner during training, as well as careful reward shaping and environment sampling. Once learned, the resulting policy can be copied onto any number of agents and naturally scales to different team sizes and world dimensions. We present results on randomized worlds with up to 1024 agents and compare success rates against state-of-the-art MAPF planners. Finally, we experimentally validate the learned policies in a hybrid simulation of a factory mockup, involving both real world and simulated robots.

关键词： Path planning for multiple mobile robots or agents deep learning in robotics and automation distributed robot systems AI-based methods factory automation

来源：评论

学校读者我要写书评

暂无评论

deep Metadata Fusion for Traffic Light to Lane Assignment

引用

IEEE robotics AND automation LETTERS 2019年第2期4卷 973-980页

作者： Langenberg, Tristan Lueddecke, Timo Woergoetter, Florentin Daimler AG Mercedes Benz Cars Res & Dev D-71063 Sindelfingen Germany Georg August Univ Gottingen Phys Inst 3 D-37077 Gottingen Germany

We present a deep metadata fusion approach that connects image data and heterogeneous metadata inside a Convolutional Neural Network (CNN). This approach enables us to assign all relevant traffic lights to their associated lanes. To achieve this, a common CNN topology is trained by down-sampled and transformed input images to predict an indication vector. The indication vector contains the column positions of all the relevant traffic lights that are associated with lanes. In parallel, we fuse prepared and adaptively weighted Metadata Feature Maps (MFM) with the convolutional feature map input of a selected convolutional layer. The results are compared to rule-based, only-metadata, and only-vision approaches. In addition, human performance of the traffic light to ego-vehicle lane assignment has been measured by a subjective test. The proposed approach outperforms all other approaches. It achieves about 93.0% average precision for a real-world dataset. In a more complex dataset, 87.1% average precision is achieved. In particular, the new approach reaches significantly higher results with 93.7% to 91.0% average accuracy for a real-world dataset in contrast to lower human performance.

关键词： Intelligent transportation systems computer vision for transportation deep learning in robotics and automation

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：