추가 적립 안내

로봇에게 지능을 부여할 수 있는 책

지능시스템은 패턴 인식과 지능제어로 나눌 수 있다. 패턴 인식 분야는 딥러닝의 개발로 정확한 영상 및 영상 인식이 가능해졌다. 지능제어는 기존의 제어 시스템에 지능을 추가하는 구조이다. 제어의 구조상 지능시스템의 온라인 학습이 요구되며 시스템에 대한 빠른 적응이 필요하다. 패턴 기반의 판별력은 오프라인으로 학습을 많이 하지만 지능제어의 경우에는 실시간으로 제어가 구현되므로 실시간 학습이 보장되어야 한다. 본 책에서는 다양한 지능제어 이론과 실제 응용에 대해 소개한다. 지능제어이론을 로봇에 적용한 실제 사례를 소개한다.

1장 적응 시스템
1.1 소개 ··································································································· 12
1.2 신호의 특성 ························································································ 15
1.3 그래디언트(기울기) 알고리즘 ······························································ 21
1.4 이산 시스템 ························································································ 26
1.5 z 변환 ································································································ 30
1.6 Wiener filter ······················································································· 36
1.7 LMS 알고리즘 ····················································································· 42
1.8 RLS 알고리즘 ······················································································ 48
1.9 Kalman 필터 ······················································································ 63
1.10 시스템 인식 모델 구조 ······································································· 72
1.11 인식 모델 ·························································································· 75
2장 적응제어
2.1 소개 ··································································································· 82
2.2 모델을 기반으로 하는 적응제어 방식 ··················································· 84
2.3 Self-tuning adaptive control ································································ 87
2.4 시스템 변수의 평가 ············································································· 88
2.5 쌍적분 시스템의 MRAC 제어의 예 ······················································· 89
2.6 로봇의 적응제어 예 ············································································· 93
2.7 필터동역학을 이용한 적응제어 ··························································· 99
3장 신경회로망
3.1 단층 퍼셉트론 넷의 소개 ·································································· 106
3.2 비선형함수 ······················································································· 108
3.3 다층 퍼셉트론 넷 ·············································································· 110
3.4. 역전파 알고리즘 ·············································································· 116
3.5. 방사형 함수 기반 신경회로망 ··························································· 141
3.6. Recurrent Network(Feedback network) ··········································· 149
3.7 XOR 패턴인식 응용 ··········································································· 157
3.8 MATLAB 명령어를 이용한 신경회로망 ··············································· 163
3.9 패턴인식 예 ······················································································ 174
3.10 제스처 패턴인식의 이륜 로봇에의 응용 ············································ 177
4장 신경회로망 제어
4.1 소개 ································································································· 184
4.2 신경회로망을 이용한 시스템 인식 ······················································ 186
4.3 신경회로망의 역모델 직접 제어 ························································· 188
4.4 시스템 자코비안 ··············································································· 193
4.5 Feedback error learning(FEL) 방식 ················································· 195
4.6 Reference Compensation Technique(RCT) 방식 ································ 198
4.7 PID 이득값 튜닝 방식 ········································································ 204
4.8 역진자-수레 시스템 제어 예 ······························································· 214
5장 로봇의 신경망 제어
5.1. 소개 ································································································ 226
5.2. 로봇의 동적 모델기반 FEL 제어 ························································ 229
5.3. 로봇의 무모델기반 FEL 제어 ···························································· 234
5.4. 로봇의 모델기반 RCT 제어 ······························································· 240
5.5. 2축 로봇의 신경망 제어 응용 ···························································· 245
5.6. 뉴로 슬라이딩 모드 제어 ·································································· 274
5.7. 뉴로 시간지연 제어 ·········································································· 295
6장 퍼지논리
6.1 소개 ································································································· 316
6.2 퍼지 셋의 정의 ·················································································· 318
6.3 소속 함수 ·························································································· 324
6.4 소속 함수의 종류 및 특성 ·································································· 325
6.5 소속 함수 만들기 ·············································································· 327
6.6 퍼지셋 이론 ······················································································ 335
6.7 퍼지관계 ··························································································· 338
6.8 퍼지추론(fuzzy implication) ······························································ 341
6.9 퍼지 합성 ·························································································· 343
6.10 퍼지 제어 논리 ················································································ 347
6.11 Fuzzy Tool Box 사용하기 ······························································· 352
7 장 퍼지 제어 시스템
7.1 소개 ································································································· 358
7.2 입력의 퍼지화 ··················································································· 360
7.3 퍼지 제어 법칙(fuzzy rule) ································································ 365
7.4 퍼지 제어 법칙 만들기 ······································································ 374
7.5 퍼지 제어기 추론 ·············································································· 381
7.6 퍼지 집합 합성 ·················································································· 389
7.7 비퍼지화 ··························································································· 391
7.8 퍼지 제어 예 ····················································································· 398
7.9 역진자 제어 예 ·················································································· 400
8장 뉴로-퍼지 제어시스템
8.1 소개 ································································································· 408
8.2 뉴로-퍼지 소속 함수 ·········································································· 410
8.3 뉴로-퍼지 제어기 구조 ······································································ 412
8.4 TSK 뉴로-퍼지 제어기 ······································································ 419
8.5 퍼지 제어의 신경망 보상 방식 ··························································· 431
8.6 MATLAB명령어로 뉴로-퍼지 구현 ······················································ 435
8.7 역진자 시스템 제어예 ········································································ 440
8.8 로봇팔 제어예 ··················································································· 445
찾아보기 ······························································································ 449

신경회로망 제어 방식의 하나인 입력 경로에 보상하는 입력보상방식(RCT: Reference compensation technique)이 있다. 입력보상방식은 자코비안 정보가 필요하지 않고 신경회로망의 출력을 입력에 보상하여 시스템의 비선형성을 제거하는 제어 방식의 일종이다. 첫째로, 제어할 대상이 PD(Proportional and derivative)제어기로 먼저 안정된 후에 보상이 되기 때문에 안정성이 높다. 역모델 제어의 가장 기본 구조인 직접 제어방식은 안정성이 취약하다. 실시간 제어일 경우에 신경회로망에서 무슨 신호가 출력되는지 보장할 수 없으므로 로봇의 움직임에 대한 안정성을 보장할 수 없다. 신경망 제어기가 PD 제어기로 미리 안정된 시스템의 역모델을 인식하므로 제안한 제어방식의 안정성을 나타낸다. 둘째로, 불확실성을 보상하는 위치가 제어하는 시스템 밖에서 이루어지기 때문에 기존에 설치되어있는 실제 시스템들의 제어기들을 수정할 필요 없이 제안한 방식을 적용하기가 쉬운 이점이 있다. 특히 무선 RCT방식은 무인항공기 또는 드론의 제어에 유용하게 사용될 수 있다.

[ 저자서문 ]
지능이란 무엇인가? 지능은 다양한 분야에서 특출하게 나타나는 모든 것을 지능이라고 할 수 있다. 가수는 노래에서 미술가는 그림에서 작곡가는 공연에서 운동선수는 운동 종목에서 각기 지능을 갖고 있다. 학문적으로 볼 때 지능은 주로 학습과 기억 그 리고 판단력과 관련된다.
지능을 판별하는 것 중 하나인 학습에 의한 판단력(decision making ability)은 다양한 방면에서 사용되고 있다. 학습에 의한 판단력을 이용하는 분야는 크게 패턴 인식 분야와 지능제어 분야이다. 패턴 인식분야로는 영상과 음성 분야에서 많이 사용되는데 많은 이미지 중에서 원하는 이미지를 찾아내는 image classifier이거나 인간의 음성을 인식하여 대화를 이끌어 가는 음성인식기 등이 있다. 최근에 딥러닝의 발달로 매우 정확한 음성인식과 영상을 접할 수 있게 되었다. 오프라인에서 학습이 이루어지므로 많은 양의 데이터를 학습하면 할수록 성능이 좋아진다. 예컨대 구글이 사용하는 슈퍼 컴퓨터의 도움으로 원하는 많은 양의 학습이 이루어지고 있다. 인식기가 학습을 할 수 있는 컴퓨터가 인공지능 컴퓨터이고 로봇이면 바로 인공지능 로봇이고 인간형 로봇이면 인공지능 사이보그가 된다.
다른 한 축은 지능제어로 기존의 제어 시스템에 지능을 추가하는 구조이다. 제어의 구조상 지능시스템의 온라인 학습이 요구되며 시스템에 대한 빠른 적응이 필요하다. 패턴 기반의 판별력은 오프라인으로 학습을 많이 하지만 지능제어의 경우에는 실시간으로 제어가 구현되므로 실시간 학습이 보장되어야 한다.
인공지능시스템 시리즈 I, II가 나온 지 벌써 어느덧 20년이 지났다. 졸작임에도 불구하고 독자들의 관심에 감사를 드린다. 이 책은 인공지능시스템 II의 지능제어라는 틀 안에서 본인이 박사과정의 연구를 수행하고 학생들을 가르치면서 얻은 연구의 결과물을 중심으로 준비하였다. 수년간 지능제어 강의를 준비한 내용과 발표했던 논문들을 학생들이 이해하기 쉽도록 쉽게 풀어서 프로그램과 함께 결과를 삽입하며 작성하였다.
이 책이 나오기까지 격려를 아끼지 않고 도와주신 캘리포니아 대학의 지도교수 Hsia 선생님께 감사드린다. 하루하루의 삶에 감사드리고, 하늘에 계신 아버지, 항상 나를 생각해 주시는 어머니, 아내와 자녀에게 감사하다는 말을 전하고 싶다. 내 강의를 듣는 학생들에게도 감사하고 실험실 학생들에게 고마움을 표시하고 싶다.
끝으로 이 책을 통하여 조금이나마 학생들의 학문적 발전에 도움이 되었으면 하는 것이 본인의 작은 바램이다.
2021년 6월
오늘도 하루에 감사하며 백마산 기슭의
실험실에서 글을 쓴다.