本文選題：自主抓取 + 深度學(xué)習(xí)�。� 參考：《哈爾濱工業(yè)大學(xué)》2017年碩士論文

【摘要】：機(jī)器人自主抓取問(wèn)題是機(jī)器人研究領(lǐng)域的一個(gè)重要問(wèn)題,過(guò)去幾十年,研究人員從不同角度展開(kāi)探索,涉及夾持器結(jié)構(gòu)設(shè)計(jì)、抓取規(guī)劃控制、多傳感器融合等,從分析式方法到數(shù)據(jù)驅(qū)動(dòng)式方法的發(fā)展軌跡可以看出,抓取研究的進(jìn)展伴隨新技術(shù)、新方法的不斷應(yīng)用。當(dāng)前,大數(shù)據(jù)、深度學(xué)習(xí)時(shí)代已經(jīng)到來(lái),將新的數(shù)據(jù)規(guī)模與新的研究方法結(jié)合能夠產(chǎn)生不可估量的效果。深度學(xué)習(xí)的產(chǎn)生,給其他研究領(lǐng)域帶來(lái)了大量啟發(fā),為此,本文在機(jī)器人自主抓取問(wèn)題上應(yīng)用深度學(xué)習(xí)技術(shù),建立抓取分類模型,實(shí)現(xiàn)抓取檢測(cè)系統(tǒng)的構(gòu)建。基于深度學(xué)習(xí)建立了抓取分類器。本文區(qū)分?jǐn)?shù)據(jù)模態(tài)之間的不同,采用數(shù)據(jù)末端融合的方法,建立了基于多模態(tài)卷積神經(jīng)網(wǎng)絡(luò)的抓取分類模型。首先,在圖像數(shù)據(jù)和深度數(shù)據(jù)上分別應(yīng)用卷積神經(jīng)網(wǎng)絡(luò)訓(xùn)練兩個(gè)抓取分類器,然后將兩個(gè)分類器作為特征提取器,分別提取樣本矩形的圖像抓取特征和深度抓取特征,最后,融合兩種抓取特征再次構(gòu)建一個(gè)頂層分類模型。聯(lián)合兩個(gè)特征提取器以及頂層分類網(wǎng),本文構(gòu)建了一個(gè)基于多模態(tài)深度學(xué)習(xí)算法的抓取分類器,實(shí)現(xiàn)了抓取矩形的精確分類。構(gòu)建了自主抓取系統(tǒng)。首先,應(yīng)用中值濾波填充Kinect深度圖像缺失,標(biāo)定Kinect獲得機(jī)器人基坐標(biāo)系下的點(diǎn)云以及對(duì)齊的圖像數(shù)據(jù),這部分內(nèi)容實(shí)現(xiàn)場(chǎng)景傳感數(shù)據(jù)的獲取。其次,基于隨機(jī)采樣一致算法提取桌面實(shí)現(xiàn)目標(biāo)分割,借助Sobel算子檢測(cè)目標(biāo)物體主方向,離散抓取五維搜索空間,這部分內(nèi)容完成抓取矩形的采樣工作。接著,在給定場(chǎng)景點(diǎn)云上剔除離群點(diǎn)并進(jìn)行均值濾波,為點(diǎn)云法向量估計(jì)做準(zhǔn)備,對(duì)檢測(cè)獲得的最優(yōu)抓取矩形映射一組最優(yōu)抓取參數(shù),這部分內(nèi)容完成抓取參數(shù)的獲取。最后,調(diào)用逆運(yùn)動(dòng)學(xué)求解服務(wù)計(jì)算Baxter機(jī)器人抓取位姿的關(guān)節(jié)角位置,驅(qū)動(dòng)Baxter到達(dá)此關(guān)節(jié)角位置實(shí)現(xiàn)對(duì)目標(biāo)物體的成功抓取。整合以上各部分構(gòu)建了自主抓取系統(tǒng)。完成了抓取分類模型的評(píng)估以及機(jī)器人抓取系統(tǒng)的實(shí)驗(yàn)研究。本文對(duì)獲得的三個(gè)抓取分類模型在數(shù)據(jù)集上進(jìn)行檢測(cè)對(duì)比分析,檢測(cè)結(jié)果表明多模態(tài)抓取分類器比單一模態(tài)的抓取分類器分類性能更優(yōu)。最后,基于多模態(tài)分類器搭建了自主抓取系統(tǒng),系統(tǒng)中Kinect實(shí)現(xiàn)場(chǎng)景數(shù)據(jù)的獲取,工作站完成最優(yōu)抓取參數(shù)的推理,Baxter機(jī)器人實(shí)現(xiàn)目標(biāo)物體的抓取等。實(shí)驗(yàn)表明,本文的抓取系統(tǒng)是可行且穩(wěn)定的。
[Abstract]:Autonomous robot capture is an important problem in the field of robot research. In the past few decades, researchers have explored from different angles, involving the structure design of gripper, grab planning control, multi-sensor fusion, etc. From the development of analytical method to data-driven method, it can be seen that the development of capture research is accompanied by new technology and new methods. At present, big data, the era of deep learning has come, combining new data scale with new research methods can produce incalculable results. The generation of deep learning has brought a lot of inspiration to other research fields. Therefore, this paper applies depth learning technology to autonomous robot grab problem, establishes grab classification model, and realizes the construction of grab detection system. A grab classifier is established based on deep learning. In this paper, the data terminal fusion method is used to distinguish the differences between data modes, and a grab classification model based on multi-modal convolution neural network is established. Firstly, two grab classifiers are trained by convolution neural network on image data and depth data, and then two classifiers are used as feature extractors to extract image grab feature and depth grab feature of sample rectangle, respectively. A top-level classification model is constructed by combining the two grab features. Combined with two feature extractors and the top-level classification network, a grab classifier based on multi-modal depth learning algorithm is constructed in this paper. An autonomous grab system is constructed. First, using median filter to fill in the missing Kinect depth image, calibrating Kinect to obtain the point cloud and aligned image data in the robot coordinate system, which realizes the acquisition of scene sensing data. Secondly, the object segmentation is realized by extracting the desktop based on the random sampling consistent algorithm, the main direction of the object is detected by Sobel operator, and the five-dimensional search space is grabbed discretely, which completes the sampling of the grab rectangle. Then, outliers are removed from the point cloud in a given scene and mean value filtering is carried out to prepare the point cloud normal vector estimation, and a set of optimal grab parameters are obtained for the detection of the optimal grab rectangle mapping, which completes the acquisition of the grab parameters. Finally, the inverse kinematics solution service is used to calculate the joint angle position of the Baxter robot, and the Baxter is driven to the joint angle position to achieve the successful capture of the target object. Integrate the above parts to build an autonomous grab system. The evaluation of grab classification model and the experimental study of robot grab system are completed. In this paper, the three grab classification models are compared with each other on the dataset. The results show that the multi-mode grab classifier is better than the single-mode grab classifier. Finally, an autonomous grab system based on multi-modal classifier is built. Kinect realizes scene data acquisition in the system, and a Baxter robot accomplishes the inference of optimal grab parameters. Experiments show that the grab system is feasible and stable.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：TP391.41;TP242

【參考文獻(xiàn)】

相關(guān)期刊論文 前4條

1 仲訓(xùn)杲;徐敏;仲訓(xùn)昱;彭俠夫;;基于多模特征深度學(xué)習(xí)的機(jī)器人抓取判別方法[J];自動(dòng)化學(xué)報(bào);2016年07期

2 盧良鋒;謝志軍;葉宏武;;基于RGB特征與深度特征融合的物體識(shí)別算法[J];計(jì)算機(jī)工程;2016年05期

3 GUO Di;SUN FuChun;LIU ChunFang;;A system of robotic grasping with experience acquisition[J];Science China(Information Sciences);2014年12期

4 秦志強(qiáng),趙錫芳,李澤湘;機(jī)器人多指抓取的力封閉判別[J];上海交通大學(xué)學(xué)報(bào);1999年07期

相關(guān)博士學(xué)位論文 前1條

1 楊揚(yáng);基于機(jī)器視覺(jué)的服務(wù)機(jī)器人智能抓取研究[D];上海交通大學(xué);2014年

相關(guān)碩士學(xué)位論文 前1條

1 李勵(lì)耘;基于高斯過(guò)程的抓取規(guī)劃方法研究[D];哈爾濱工業(yè)大學(xué);2015年

，

本文編號(hào)：2068608