scholarly journals A Pushing-Grasping Collaborative Method Based on Deep Q-Network Algorithm in Dual Viewpoints

2021 ◽  
Author(s):  
Gang Peng ◽  
Jinhu Liao ◽  
Shangbin Guan ◽  
Jin Yang ◽  
Xinde Li
Keyword(s):  
Point Clouds ◽  
Intelligent Manufacturing ◽  
Well Performance ◽  
Reward Function ◽  
Network Algorithm ◽  
Novel Objects ◽  
Dual Perspective ◽  
Capture Method ◽  
Speed Up ◽  
Robot Grasping

Abstract In the field of intelligent manufacturing, robot grasping and sorting is an important content. However, in the traditional 2D camera-based robotic arm grasping method, the grasping efficiency is low and the grasping accuracy is low when facing the scene of stacking and occlusion. Insufficiency and other issues, a dual perspective-based deep reinforcement learning promotion and capture method is proposed. In this case, a novel method of pushing-grasping collaborative based on the deep Q-network in dual viewpoints is proposed in this paper. This method adopts an improved deep Q-network algorithm, with an RGB-D camera to obtain the information of objects’ RGB images and point clouds from two viewpoints, and combines the pushing and grasping actions, so that the trained manipulator can make the scenes better for grasping, so that it can perform well in more complicated grasping scenes. What’s more, we improved the reward function of the deep Q-network and propose the piecewise reward function to speed up the convergence of the deep Q-network. We trained different models and tried different methods in the V-REP simulation environment, and it drew a conclusion that the method proposed in this paper converges quickly and the success rate of grasping objects in unstructured scenes raises up to 83.5\%. Besides, it shows the generalization ability and well performance when novel objects appear in the scenes that the manipulator has never grasped before.

scholarly journals Automatic Calibration of a Two-Axis Rotary Table for 3D Scanning Purposes

Sensors ◽  
2020 ◽  
Vol 20 (24) ◽  
pp. 7107
Author(s):  
Livio Bisogni ◽  
Ramtin Mollaiyan ◽  
Matteo Pettinari ◽  
Paolo Neri ◽  
Marco Gabiccini
Keyword(s):  
Point Clouds ◽  
3D Scanning ◽  
Optimal Number ◽  
Acquisition Time ◽  
Complex Geometries ◽  
Automatic Calibration ◽  
Camera System ◽  
Reprojection Error ◽  
Speed Up ◽  
Manual Registration

Rotary tables are often used to speed up the acquisition time during the 3D scanning of complex geometries. In order to avoid manual registration of the point clouds acquired with different orientations, automatic algorithms to compensate the rotation were developed. Alternatively, a proper calibration of the rotary axis with respect to the camera system is needed. Several methods are available in the literature, but they only consider a single-axis calibration. In this paper, a method for the simultaneous calibration of both axes of the table is proposed. A checkerboard is attached to the table, and several images with different poses are acquired. An optimization algorithm is then setup to determine the orientation and the locations of the two axes. A metric to assess the calibration quality was also defined by computing the average mean reprojection error. This metric is used to investigate the optimal number and distribution of the calibration poses, demonstrating that the optimum calibration results are achieved when a wider dispersion of the calibration poses is adopted.

scholarly journals PARAMETRIC MODELLING APPROACH TO RECONSTRUCTING ARCHITECTURAL INDOOR SPACES FROM POINT CLOUDS

2021 ◽  
Vol XLIII-B4-2021 ◽  
pp. 251-257
Author(s):  
W. Wahbeh
Keyword(s):  
Programming Languages ◽  
Point Clouds ◽  
Visual Programming ◽  
Parametric Modelling ◽  
Architectural Practice ◽  
Industry Foundation Classes ◽  
Interior Spaces ◽  
Speed Up ◽  
Indoor Spaces ◽  
Modelling Approach

Abstract. In this paper, some outcomes of a research project which aims to introduce automation to speed up modelling of architectural spaces based on point clouds are presented. The main objective of the research is to replace some manual parametric modelling steps with automatic processes to obtain editable models in BIM-ready software and not to generate non-parametric IFC (Industry Foundation Classes) models. An approach of automation using visual programming for interior wall modelling based on point clouds is presented. The pipeline and the different concepts represented in this paper are applicable using different programming languages but here the use of Rhinoceros as a modelling software and its open-source visual programming extension "Grasshopper" is intentional as it is in common use for parametric modelling and generative design in architectural practice. In this research, it is assumed that there is a predominance of three mutually orthogonal directions of the walls in the interior spaces to be analysed, which is the case of most indoor spaces.

scholarly journals A VOXEL-BASED METADATA STRUCTURE FOR CHANGE DETECTION IN POINT CLOUDS OF LARGE-SCALE URBAN AREAS

2018 ◽  
Vol IV-2 ◽  
pp. 97-104
Author(s):  
J. Gehrung ◽  
M. Hebel ◽  
M. Arens ◽  
U. Stilla
Keyword(s):  
Change Detection ◽  
Urban Areas ◽  
Laser Scanning ◽  
Large Scale ◽  
Point Clouds ◽  
Urban Environments ◽  
Detection Methods ◽  
Planar Surfaces ◽  
Occupancy Grids ◽  
Speed Up

Mobile laser scanning has not only the potential to create detailed representations of urban environments, but also to determine changes up to a very detailed level. An environment representation for change detection in large scale urban environments based on point clouds has drawbacks in terms of memory scalability. Volumes, however, are a promising building block for memory efficient change detection methods. The challenge of working with 3D occupancy grids is that the usual raycasting-based methods applied for their generation lead to artifacts caused by the traversal of unfavorable discretized space. These artifacts have the potential to distort the state of voxels in close proximity to planar structures. In this work we propose a raycasting approach that utilizes knowledge about planar surfaces to completely prevent this kind of artifacts. To demonstrate the capabilities of our approach, a method for the iterative volumetric approximation of point clouds that allows to speed up the raycasting by 36 percent is proposed.

Simulation and deep learning on point clouds for robot grasping

2021 ◽  
Vol ahead-of-print (ahead-of-print) ◽  
Author(s):  
Zhengtuo Wang ◽  
Yuetong Xu ◽  
Guanhua Xu ◽  
Jianzhong Fu ◽  
Jiongyan Yu ◽  
...  
Keyword(s):  
Deep Learning ◽  
Pose Estimation ◽  
Point Clouds ◽  
Estimation Algorithm ◽  
Training Data ◽  
Learning Method ◽  
Data Set ◽  
Content Type ◽  
Experimental Platform ◽  
Robot Grasping

Purpose In this work, the authors aim to provide a set of convenient methods for generating training data, and then develop a deep learning method based on point clouds to estimate the pose of target for robot grasping. Design/methodology/approach This work presents a deep learning method PointSimGrasp on point clouds for robot grasping. In PointSimGrasp, a point cloud emulator is introduced to generate training data and a pose estimation algorithm, which, based on deep learning, is designed. After trained with the emulation data set, the pose estimation algorithm could estimate the pose of target. Findings In experiment part, an experimental platform is built, which contains a six-axis industrial robot, a binocular structured-light sensor and a base platform with adjustable inclination. A data set that contains three subsets is set up on the experimental platform. After trained with the emulation data set, the PointSimGrasp is tested on the experimental data set, and an average translation error of about 2–3 mm and an average rotation error of about 2–5 degrees are obtained. Originality/value The contributions are as follows: first, a deep learning method on point clouds is proposed to estimate 6D pose of target; second, a convenient training method for pose estimation algorithm is presented and a point cloud emulator is introduced to generate training data; finally, an experimental platform is built, and the PointSimGrasp is tested on the platform.

scholarly journals Grasp Pose Detection in Point Clouds

2017 ◽  
Vol 36 (13-14) ◽  
pp. 1455-1473 ◽  
Author(s):  
Andreas ten Pas ◽  
Marcus Gualtieri ◽  
Kate Saenko ◽  
Robert Platt
Keyword(s):  
Point Cloud ◽  
Point Clouds ◽  
Sensor Data ◽  
Detection Methods ◽  
Cad Model ◽  
Success Rates ◽  
Pose Detection ◽  
Novel Objects ◽  
Partially Occluded ◽  
Robotic Grasp

Recently, a number of grasp detection methods have been proposed that can be used to localize robotic grasp configurations directly from sensor data without estimating object pose. The underlying idea is to treat grasp perception analogously to object detection in computer vision. These methods take as input a noisy and partially occluded RGBD image or point cloud and produce as output pose estimates of viable grasps, without assuming a known CAD model of the object. Although these methods generalize grasp knowledge to new objects well, they have not yet been demonstrated to be reliable enough for wide use. Many grasp detection methods achieve grasp success rates (grasp successes as a fraction of the total number of grasp attempts) between 75% and 95% for novel objects presented in isolation or in light clutter. Not only are these success rates too low for practical grasping applications, but the light clutter scenarios that are evaluated often do not reflect the realities of real-world grasping. This paper proposes a number of innovations that together result in an improvement in grasp detection performance. The specific improvement in performance due to each of our contributions is quantitatively measured either in simulation or on robotic hardware. Ultimately, we report a series of robotic experiments that average a 93% end-to-end grasp success rate for novel objects presented in dense clutter.

scholarly journals POINT CLOUD MANAGEMENT THROUGH THE REALIZATION OF THE INTELLIGENT CLOUD VIEWER SOFTWARE

2017 ◽  
Vol XLII-5/W1 ◽  
pp. 105-112 ◽  
Author(s):  
D. Costantino ◽  
M. G. Angelini ◽  
F. Settembrini
Keyword(s):  
Computational Geometry ◽  
Point Cloud ◽  
High Performance ◽  
Laser Scanner ◽  
Ground Level ◽  
Point Clouds ◽  
Open Inventor ◽  
Speed Up ◽  
High Performance Systems ◽  
Made In

The paper presents a software dedicated to the elaboration of point clouds, called <i>Intelligent Cloud Viewer</i> (ICV), made in-house by AESEI software (Spin-Off of Politecnico di Bari), allowing to view point cloud of several tens of millions of points, also on of “no” very high performance systems. The elaborations are carried out on the whole point cloud and managed by means of the display only part of it in order to speed up rendering. It is designed for 64-bit Windows and is fully written in C ++ and integrates different specialized modules for computer graphics (Open Inventor by SGI, Silicon Graphics Inc), maths (BLAS, EIGEN), computational geometry (CGAL, <i>Computational Geometry Algorithms Library</i>), registration and advanced algorithms for point clouds (PCL, <i>Point Cloud Library</i>), advanced data structures (BOOST, <i>Basic Object Oriented Supporting Tools</i>), etc. ICV incorporates a number of features such as, for example, cropping, transformation and georeferencing, matching, registration, decimation, sections, distances calculation between clouds, etc. It has been tested on photographic and TLS (<i>Terrestrial Laser Scanner</i>) data, obtaining satisfactory results. The potentialities of the software have been tested by carrying out the photogrammetric survey of the Castel del Monte which was already available in previous laser scanner survey made from the ground by the same authors. For the aerophotogrammetric survey has been adopted a flight height of approximately 1000ft AGL (<i>Above Ground Level</i>) and, overall, have been acquired over 800 photos in just over 15 minutes, with a covering not less than 80%, the planned speed of about 90 knots.

scholarly journals Fast reinforcement learning with generalized policy updates

2020 ◽  
Vol 117 (48) ◽  
pp. 30079-30087 ◽  
Author(s):  
André Barreto ◽  
Shaobo Hou ◽  
Diana Borsa ◽  
David Silver ◽  
Doina Precup
Keyword(s):  
Reinforcement Learning ◽  
Divide And Conquer ◽  
Problem Decomposition ◽  
Sequential Decision ◽  
Learning Problem ◽  
Reward Function ◽  
Speed Up ◽  
Complex Decision ◽  
Reward Functions ◽  
Do So

The combination of reinforcement learning with deep learning is a promising approach to tackle important sequential decision-making problems that are currently intractable. One obstacle to overcome is the amount of data needed by learning systems of this type. In this article, we propose to address this issue through a divide-and-conquer approach. We argue that complex decision problems can be naturally decomposed into multiple tasks that unfold in sequence or in parallel. By associating each task with a reward function, this problem decomposition can be seamlessly accommodated within the standard reinforcement-learning formalism. The specific way we do so is through a generalization of two fundamental operations in reinforcement learning: policy improvement and policy evaluation. The generalized version of these operations allow one to leverage the solution of some tasks to speed up the solution of others. If the reward function of a task can be well approximated as a linear combination of the reward functions of tasks previously solved, we can reduce a reinforcement-learning problem to a simpler linear regression. When this is not the case, the agent can still exploit the task solutions by using them to interact with and learn about the environment. Both strategies considerably reduce the amount of data needed to solve a reinforcement-learning problem.

scholarly journals COMPARISON OF SINGLE AND MULTI-SCALE METHOD FOR LEAF AND WOOD POINTS CLASSIFICATION FROM TERRESTRIAL LASER SCANNING DATA

2018 ◽  
Vol IV-3 ◽  
pp. 217-223
Author(s):  
Hongqiang Wei ◽  
Guiyun Zhou ◽  
Junjie Zhou
Keyword(s):  
Classification Accuracy ◽  
Laser Scanning ◽  
Terrestrial Laser Scanning ◽  
Point Clouds ◽  
Multi Scale ◽  
Scale Method ◽  
Oak Trees ◽  
Single Scale ◽  
Speed Up ◽  
Salient Features

The classification of leaf and wood points is an essential preprocessing step for extracting inventory measurements and canopy characterization of trees from the terrestrial laser scanning (TLS) data. The geometry-based approach is one of the widely used classification method. In the geometry-based method, it is common practice to extract salient features at one single scale before the features are used for classification. It remains unclear how different scale(s) used affect the classification accuracy and efficiency. To assess the scale effect on the classification accuracy and efficiency, we extracted the single-scale and multi-scale salient features from the point clouds of two oak trees of different sizes and conducted the classification on leaf and wood. Our experimental results show that the balanced accuracy of the multi-scale method is higher than the average balanced accuracy of the single-scale method by about 10&amp;thinsp;% for both trees. The average speed-up ratio of single scale classifiers over multi-scale classifier for each tree is higher than 30.

Learning ambidextrous robot grasping policies

2019 ◽  
Vol 4 (26) ◽  
pp. eaau4984 ◽  
Author(s):  
Jeffrey Mahler ◽  
Matthew Matl ◽  
Vishal Satish ◽  
Michael Danielczuk ◽  
Bill DeRose ◽  
...  
Keyword(s):  
Three Dimensional ◽  
Service Robots ◽  
Grand Challenge ◽  
Diverse Range ◽  
Three Dimensional Objects ◽  
Depth Images ◽  
Suction Cup ◽  
Novel Objects ◽  
Robot Grasping ◽  
Synthetic Datasets

Universal picking (UP), or reliable robot grasping of a diverse range of novel objects from heaps, is a grand challenge for e-commerce order fulfillment, manufacturing, inspection, and home service robots. Optimizing the rate, reliability, and range of UP is difficult due to inherent uncertainty in sensing, control, and contact physics. This paper explores “ambidextrous” robot grasping, where two or more heterogeneous grippers are used. We present Dexterity Network (Dex-Net) 4.0, a substantial extension to previous versions of Dex-Net that learns policies for a given set of grippers by training on synthetic datasets using domain randomization with analytic models of physics and geometry. We train policies for a parallel-jaw and a vacuum-based suction cup gripper on 5 million synthetic depth images, grasps, and rewards generated from heaps of three-dimensional objects. On a physical robot with two grippers, the Dex-Net 4.0 policy consistently clears bins of up to 25 novel objects with reliability greater than 95% at a rate of more than 300 mean picks per hour.

Sign in / Sign up

Export Citation Format

Share Document