基于多模特征深度学习的机器人抓取判别方法

针对智能机器人抓取判别问题,研究多模特征深度学习与融合方法.该方法将测试特征分布偏离训练特征视为一类噪化,引入带稀疏约束的降噪自动编码（Denoising auto-encoding, DAE）,实现网络权值学习;并以叠层融合策略,获取初始多模特征的深层抽象表达,两种手段相结合旨在提高深度网络的鲁棒性和抓取判别精确性.实验采用深度摄像机与6自由度工业机器人组建测试平台,对不同类别目标进行在线对比实验.结果表明,设计的多模特征深度学习依据人的抓取习惯,实现最优抓取判别,并且机器人成功实施抓取定位,研究方法对新目标具备良好的抓取判别能力.     

Grasp programming by demonstration in virtual reality with automatic environment reconstruction

A virtual reality system enabling high-level programming of robot grasps is described. The system is designed to support programming by demonstration (PbD), an approach aimed at simplifying robot programming and empowering even unexperienced users with the ability to easily transfer knowledge to a robotic system. Programming robot grasps from human demonstrations requires an analysis phase, comprising learning and classification of human grasps, as well as a synthesis phase, where an appropriate human-demonstrated grasp is imitated and adapted to a specific robotic device and object to be grasped. The virtual reality system described in this paper supports both phases, thereby enabling end-to-end imitation-based programming of robot grasps. Moreover, as in the PbD approach robot environment interactions are no longer explicitly programmed, the system includes a method for automatic environment reconstruction that relieves the designer from manually editing the pose of the objects in the scene and enables intelligent manipulation. A workspace modeling technique based on monocular vision and computation of edge-face graphs is proposed. The modeling algorithm works in real time and supports registration of multiple views. Object recognition and workspace reconstruction features, along with grasp analysis and synthesis, have been tested in simulated tasks involving 3D user interaction and programming of assembly operations. Experiments reported in the paper assess the capabilities of the three main components of the system: the grasp recognizer, the vision-based environment modeling system, and the grasp synthesizer.     

Implementation of a Gaussian process-based machine learning grasp predictor

With the goal of advancing the state of automatic robotic grasping, we present a novel approach that combines machine learning techniques and physical validation on a robotic platform to develop a comprehensive grasp predictor. After collecting a large grasp sample set (522 grasps), we first conduct a statistical analysis of the predictive ability of grasp quality metrics that are commonly used in the robotics literature. We then apply principal component analysis and Gaussian process (GP) algorithms on the grasp metrics that are discriminative to build a classifier, validate its performance, and compare the results to existing grasp planners. The key findings are as follows: (i) several of the existing grasp metrics are weak predictors of grasp quality when implemented on a robotic platform; (ii) the GP-based classifier significantly improves grasp prediction by combining multiple grasp metrics to increase true positive classification at low false positive rates; (iii) The GP classifier can be used generate new grasps to improve bad grasp samples by performing a local search to find neighboring grasps which have improved contact points and higher success rate.     

基于多约束条件的机器人抓取策略学习方法

针对机器人在多类别物体不同任务下的抓取决策问题,提出基于多约束条件的抓取策略学习方法.该方法以抓取对象特征和抓取任务属性为机器人抓取策略约束,通过映射人类抓取习惯规划抓取模式,并采用物体方向包围盒(OBB)建立机器人抓取规则,建立多约束条件的抓取模型.利用深度径向基(DRBF)网络模型结合减聚类算法(SCM)实现抓取策略的学习,两种算法的结合旨在提高学习鲁棒性与精确性.搭建以Refiex 1型灵巧手和AUBO六自由度机械臂组成的实验平台,对多类别物体进行抓取实验.实验结果表明,所提出方法使机器人有效学习到对多物体不同任务的最优抓取策略,具有良好的抓取决策能力.     

Hierarchical reinforcement learning of multiple grasping strategies with human instructions

Grasping is an essential component for robotic manipulation and has been investigated for decades. Prior work on grasping often assumes that a sufficient amount of training data is available for learning and planning robotic grasps. However, constructing such an exhaustive training dataset is very challenging in practice, and it is desirable that a robotic system can autonomously learn and improves its grasping strategy. Although recent work has presented autonomous data collection through trial and error, such methods are often limited to a single grasp type, e.g. vertical pinch grasp. To address these issues, we present a hierarchical policy search approach for learning multiple grasping strategies. To leverage human knowledge, multiple grasping strategies are initialized with human demonstrations. In addition, a database of grasping motions and point clouds of objects is also autonomously built upon a set of grasps given by a user. The problem of selecting the grasp location and grasp policy is formulated as a bandit problem in our framework. We applied our reinforcement learning to grasping both rigid and deformable objects. The experimental results show that our framework autonomously learns and improves its performance through trial and error and can grasp previously unseen objects with a high accuracy.     

基于人类演示视频的机器人指令生成框架

为了提高机器人学习技能的能力,免除人工示教过程,本文基于对无特殊标记的人类演示视频的观察,提出了一种基于序列到序列模式的机器人指令自动生成框架。首先,使用Mask R-CNN（区域卷积神经网络）来缩小操作区域的范围,并采用双流I3D网络（膨胀3D卷积网络）从视频中提取光流特征和RGB特征;其次,引入双向LSTM（长短期记忆）网络从先前提取的特征中获取上下文信息;最后,使用自我注意力机制和全局注意力机制,学习视频帧序列和命令序列的关联性,序列到序列模型最终输出机器人的命令。在扩展后的MPII烹饪活动2数据集和IIT-V2C数据集上进行了大量的实验,与现有的方法进行比较,本文提出的方法在BLEU_4（0.705）和METEOR（0.462）等指标上达到目前最先进性能水平。结果表明,该方法能够从人类演示视频中学习操作任务。此外,本框架成功应用于Baxter机器人。     

基于深度神经网络的多指灵巧手抓取手势优化

基于深度神经网络模型,提出了一种适用于多指灵巧手的抓取手势优化方法。首先,在仿真环境下构建了一个抓取数据集,并在此基础上训练了一个卷积神经网络,依据目标物体单目视觉信息和多指灵巧手抓取位形来预测抓取质量函数,由此可以将多指灵巧手的抓取规划问题转化为使抓取质量最大化的优化问题,进一步,基于深度神经网络中的反向传播和梯度上升算法实现多指灵巧手抓取手势的迭代与优化。在仿真环境中,比较该网络和仿真平台对同一抓取位形的抓取质量评估结果,再利用所提出的优化方法对随机搜索到的初始手势进行优化,比较优化前后手势的力封闭指标。最后,在实际机器人平台上验证本文方法的优化效果,结果表明,本文方法对未知物体的抓取成功率在80%以上,对于失败的抓取,优化后成功的比例达到90%。     

Towards cognitive grasping: modeling of unknown objects and its corresponding grasp types

This paper describes an intuitive approach for a cognitive grasp of a robot. The cognitive grasp means the chain of processes that make a robot to learn and execute a grasping method for unknown objects like a human. In the learning step, a robot looks around a target object to estimate the 3D shape and understands the grasp type for the object through a human demonstration. In the execution step, the robot correlates an unknown object to one of known grasp types by comparing the shape similarity of the target object based on previously learned models. For this cognitive grasp, we mainly deal with two functionalities such as reconstructing an unknown 3D object and classifying the object by grasp types. In the experiment, we evaluate the performance of object classification according to the grasp types for 20 objects via human demonstration.     

机器人多指手灵巧抓持规划

抓持规划是机器人灵巧手要完成预期任务所面临的一个重要问题．本文采用主从操作方式进行灵巧手的指尖抓持规划，由人手决定抓持接触点的位置， 灵巧手通过调整其手掌的位置和姿态保证各手指在人手指定的位置上抓持物体．根据灵巧手的操作特点，提出以关节灵活度来描述关节运动各向同性的能力，并据此定义灵巧手操作灵活度，作为灵巧手抓持位形性能的评价指标．以最大操作灵活度作为优化目标函数，寻求最优的抓持性能．同时，借鉴人手的抓持经验，通过主从操作方式，建立从人手到灵巧手的运动映射关系，从而为手掌位置优化问题提供合理的初值．仿真实验结果说明了文中方法的有效性．     

Recognition of human grasps by time-clustering and fuzzy modeling

In this paper, we address the problem of recognition of human grasps for five-fingered robotic hands and industrial robots in the context of programming-by-demonstration. The robot is instructed by a human operator wearing a data glove capturing the hand poses. For a number of human grasps, the corresponding fingertip trajectories are modeled in time and space by fuzzy clustering and Takagi–Sugeno (TS) modeling. This so-called time-clustering leads to grasp models using time as an input parameter and fingertip positions as outputs. For a sequence of grasps, the control system of the robot hand identifies the grasp segments, classifies the grasps and generates the sequence of grasps shown before. For this purpose, each grasp is correlated with a training sequence. By means of a hybrid fuzzy model, the demonstrated grasp sequence can be reconstructed.     

基于未知物体三维点云特征的机器人六自由度抓取

针对非结构化环境中任意位姿的未知物体,提出了一种基于点云特征的机器人六自由度抓取位姿检测方法,以解决直接从点云中获取目标抓取位姿的难题.首先,根据点云的基本几何信息生成抓取候选,并通过力平衡等方法优化这些候选;然后,利用可直接处理点云的卷积神经网络ConvPoint评估样本,得分最高的抓取将被执行,其中抓取位姿采样和评估网络都是以原始点云作为输入;最后,利用仿真和实际抓取实验进行测试.结果表明,该方法在常用对象上实现了88.33%的抓取成功率,并可以有效地拓展到抓取其他形状的未知物体.     

基于神经网络的机器人自学习控制器

提出一种神经网络与PID控制相结合的机器人自学习控制器.为加快神经网络的 学习收敛性,研究了有效的优化学习算法.以两关节机器人为对象的仿真表明,该控制器使机 器人跟踪希望轨迹,其系统响应、跟踪精度和鲁棒性优于常规的控制策略.     

Learning of grasp selection based on shape-templates

The ability to grasp unknown objects still remains an unsolved problem in the robotics community. One of the challenges is to choose an appropriate grasp configuration, i.e., the 6D pose of the hand relative to the object and its finger configuration. In this paper, we introduce an algorithm that is based on the assumption that similarly shaped objects can be grasped in a similar way. It is able to synthesize good grasp poses for unknown objects by finding the best matching object shape templates associated with previously demonstrated grasps. The grasp selection algorithm is able to improve over time by using the information of previous grasp attempts to adapt the ranking of the templates to new situations. We tested our approach on two different platforms, the Willow Garage PR2 and the Barrett WAM robot, which have very different hand kinematics. Furthermore, we compared our algorithm with other grasp planners and demonstrated its superior performance. The results presented in this paper show that the algorithm is able to find good grasp configurations for a large set of unknown objects from a relatively small set of demonstrations, and does improve its performance over time.     

An overview of 3D object grasp synthesis algorithms

This overview presents computational algorithms for generating 3D object grasps with autonomous multi-fingered robotic hands. Robotic grasping has been an active research subject for decades, and a great deal of effort has been spent on grasp synthesis algorithms. Existing papers focus on reviewing the mechanics of grasping and the finger–object contact interactions Bicchi and Kumar (2000) [12] or robot hand design and their control Al-Gallaf et al. (1993) [70]. Robot grasp synthesis algorithms have been reviewed in Shimoga (1996) [71], but since then an important progress has been made toward applying learning techniques to the grasping problem. This overview focuses on analytical as well as empirical grasp synthesis approaches.     

A human-like real-time grasp synthesis method for humanoid robot hands

This paper addresses a real-time grasp synthesis of multi-fingered robot hands to find grasp configurations which satisfy the force closure condition of arbitrary shaped objects. We propose a fast and efficient grasp synthesis algorithm for planar polygonal objects, which yields the contact locations on a given polygonal object to obtain a force closure grasp by a multi-fingered robot hand. For an optimum grasp and real-time computation, we develop the preference and the hibernation process and assign the physical constraints of a humanoid hand to the motion of each finger. The preferences consist of each sublayer reflecting the primitive preference similar to the conditional behaviors of humans for given objectives and their arrangements are adjusted by the heuristics of human grasping. The proposed method reduces the computational time significantly at the sacrifice of global optimality, and enables grasp posture to be changeable within 2-finger and 3-finger grasp. The performance of the presented algorithm is evaluated via simulation studies to obtain the force-closure grasps of polygonal objects with fingertip grasps. The architecture suggested is verified through experimental implementation to our developed robot hand system by solving 2- or 3-finger grasp synthesis.     

A 3D shape segmentation approach for robot grasping by parts

Neuro-psychological findings have shown that human perception of objects is based on part decomposition. Most objects are made of multiple parts which are likely to be the entities actually involved in grasp affordances. Therefore, automatic object recognition and robot grasping should take advantage from 3D shape segmentation. This paper presents an approach toward planning robot grasps across similar objects by part correspondence. The novelty of the method lies in the topological decomposition of objects that enables high-level semantic grasp planning.In particular, given a 3D model of an object, the representation is initially segmented by computing its Reeb graph. Then, automatic object recognition and part annotation are performed by applying a shape retrieval algorithm. After the recognition phase, queries are accepted for planning grasps on individual parts of the object. Finally, a robot grasp planner is invoked for finding stable grasps on the selected part of the object. Grasps are evaluated according to a widely used quality measure. Experiments performed in a simulated environment on a reasonably large dataset show the potential of topological segmentation to highlight candidate parts suitable for grasping.     

Synergy-based affordance learning for robotic grasping

In this paper, we present an affordance learning system for robotic grasping. The system involves three important aspects: the affordance memory, synergy-based exploration, and a grasping control strategy using local sensor feedback. The affordance memory is modeled with a modified growing neural gas network that allows affordances to be learned quickly from a small dataset of human grasping and object features. After being trained offline, the affordance memory is used in the system to generate online motor commands for reaching and grasping control of the robot. When grasping new objects, the system can explore various grasp postures efficiently in the low dimensional synergy space because the synergies automatically avoid abnormal postures that are more likely to lead to failed grasps. Experimental results demonstrated that the affordance memory can generalize to grasp new objects and predict the effect of the grasp (i.e., the tactile patterns).     

Shape Morphing-Based Control of Robotic Visual Servoing

We present an approach for controlling robotic interactions with objects, using synthetic images generated by morphing shapes. In particular, we attempt the problem of positioning an eye-in-hand robotic system with respect to objects in the workspace for grasping and manipulation. In our formulation, the grasp position (and consequently the approach trajectory of the manipulator), varies with each object. The proposed solution to the problem consists of two parts. First, based on a model-based object recognition framework, images of the objects taken at the desired grasp pose are stored in a database. The recognition and identification of the grasp position for an unknown input object (selected from the family of recognizable objects) occurs by morphing its contour to the templates in the database and using the virtual energy spent during the morph as a dissimilarity measure. In the second step, the images synthesized during the morph are used to guide the eye-in-hand system and execute the grasp. The proposed method requires minimal calibration of the system. Furthermore, it conjoins techniques from shape recognition, computer graphics, and vision-based robot control in a unified engineering amework. Potential applications range from recognition and positioning with respect to partially-occluded or deformable objects to planning robotic grasping based on human demonstration.     

Efficient models for grasp planning with a multi-fingered hand

This paper presents a simple grasp planning method for a multi-fingered hand. Its purpose is to compute a context-independent and dense set or list of grasps, instead of just a small set of grasps regarded as optimal with respect to a given criterion. By context-independent, we mean that only the robot hand and the object to grasp are considered. The environment and the position of the robot base with respect to the object are considered in a further stage. Such a dense set can be computed offline and then used to let the robot quickly choose a grasp adapted to a specific situation. This can be useful for manipulation planning of pick-and-place tasks. Another application is human–robot interaction when the human and robot have to hand over objects to each other. If human and robot have to work together with a predefined set of objects, grasp lists can be employed to allow a fast interaction.The proposed method uses a dense sampling of the possible hand approaches based on a simple but efficient shape feature. As this leads to many finger inverse kinematics tests, hierarchical data structures are employed to reduce the computation times. The data structures allow a fast determination of the points where the fingers can realize a contact with the object surface. The grasps are ranked according to a grasp quality criterion so that the robot will first parse the list from best to worse quality grasps, until it finds a grasp that is valid for a particular situation.     

Learning robotic hand-eye coordination through a developmental constraint driven approach

The skill of robotic hand-eye coordination not only helps robots to deal with real time environment,but also afects the fundamental framework of robotic cognition.A number of approaches have been developed in the literature for construction of the robotic hand-eye coordination.However,several important features within infant developmental procedure have not been introduced into such approaches.This paper proposes a new method for robotic hand-eye coordination by imitating the developmental progress of human infants.The work employs a brain-like neural network system inspired by infant brain structure to learn hand-eye coordination,and adopts a developmental mechanism from psychology to drive the robot.The entire learning procedure is driven by developmental constraint: The robot starts to act under fully constrained conditions,when the robot learning system becomes stable,a new constraint is assigned to the robot.After that,the robot needs to act with this new condition again.When all the contained conditions have been overcome,the robot is able to obtain hand-eye coordination ability.The work is supported by experimental evaluation,which shows that the new approach is able to drive the robot to learn autonomously,and make the robot also exhibit developmental progress similar to human infants.