검색 상세

머니퓰레이터와 시뮬레이터를 이용한 앙상블 학습 기반의 물체 파지

  • 나용호 (대학원 메카트로닉스협동과정 메카트로닉스전공)

초록/요약

Object grasping based on deep learning using a robot manipulator has been discussed in this study. The manipulator trains the pose of the gripper according to their shapes and poses in order to grasp various objects. A large amount of training data is required for good performance in deep learning. In general, training data is collected using an actual robot. However, this causes various problems such as a cost for operating the robot and securing the safety of a robot operator. Therefore, we propose a method to collect a large amount of training data using a simulator as well as an actual robot. To this end, the simulation system is composed of a manipulator, a 2-finger gripper and the work environment. Using a simulator, grasping the objects and collecting the training data are conducted repeatedly. The convolutional neural network is used for training in which its input is the RGB image of the object and its output is the pose of the gripper. Furthermore, ensemble learning is used to combine real data and simulation data. Through the proposed method, various types of objects could be grasped using a manipulator and a large amount of data could be efficiently collected through the simulator. In addition, through the ensemble learning method, we were able to improve the training performance using a large amount of training data efficiently.

more

목차


제 1장 서 론 ………………………………………………………………………… 1
1.1 연구 배경 및 동기 …………………………………………………………………1
1.2 연구 동향 ……………………………………………………………..……………2
1.3 연구의 목표 및 방법 ………………………………………………………………3

제 2장 학습을 위한 데이터 수집 ………………………………………………… 5
2.1 데이터 수집 환경 ……………………………………………………..……………5
2.2 시뮬레이터를 이용한 데이터 수집 …………………………………………….…8
2.3 실증 데이터 수집 ………………………………………………………………….13

제 3장 물체 파지를 위한 앙상블 학습 …………………………………………. 15
3.1 네트워크를 통한 학습 ....………………………………………………………….15
3.2 앙상블 학습 .………………………………………………………………….……21


제 4장 물체 파지 방법 학습의 실험 및 검증 …………………….…………… 26
4.1 학습 데이터 수집의 결과 ..………………………………………………………26
4.2 실험 환경 ………………………………………………………………………….27
4.3 단일 분류기를 이용한 학습 결과 ……………………………………………….28
4.4 앙상블 학습 기반의 학습 결과 ………………………..………………………...31

제 5장 결 론 …………………………………………………………………… 33

참고 문헌 ……………………………………………………………………… 34

more
반출 Meta View 목록