2 i C 2 i Research on Research on C Intelligent Transportation - - PowerPoint PPT Presentation

▶
2 i
SMART_READER_LITE
LIVE PREVIEW

2 i C 2 i Research on Research on C Intelligent Transportation - - PowerPoint PPT Presentation

Oct 27, 2023 339 likes •703 views

2 i C 2 i Research on Research on C Intelligent Transportation Intelligent Transportation Michel Pasquier Michel Pasquier 2 i C 2 i ) ( C Centre for Computational Intelligence ( ) Centre for Computational Intelligence Nanyang Technological

slide-1
SLIDE 1

C C2

2i

i Research on Research on Intelligent Transportation Intelligent Transportation

Michel Pasquier Michel Pasquier

Centre for Computational Intelligence Centre for Computational Intelligence ( (C

C2

2i

i)

) Nanyang Technological University Nanyang Technological University

School of Computer Engineering, Blk N4 #2A School of Computer Engineering, Blk N4 #2A-

  • 32,

32, Nanyang Avenue, Singapore 639798. Nanyang Avenue, Singapore 639798. E E-

  • mail: pasquier@pmail.ntu.edu.sg

mail: pasquier@pmail.ntu.edu.sg http://www.c2i.ntu.edu.sg/ http://www.c2i.ntu.edu.sg/

slide-2
SLIDE 2

Centre for Computational Intelligence C2i

Overview Overview

  • Research at

Research at C C2

2i

i

  • Hybrid neuro

Hybrid neuro-

  • cognitive systems

cognitive systems

  • GenSoYager fuzzy neural network

GenSoYager fuzzy neural network

  • Applications related to transportation

Applications related to transportation

  • Intelligent vehicles and driver modelling

Intelligent vehicles and driver modelling

  • Automated driver training methodology

Automated driver training methodology

  • Examples of acquired driving skills

Examples of acquired driving skills

  • Conclusion and future work

Conclusion and future work

slide-3
SLIDE 3

Centre for Computational Intelligence C2i

Overview of Overview of C C2

2i

i

  • Personnel

Personnel

– 10 faculty members, 1 RF, 2 technicians – 20 PhD and 7 MEng students – 9 PhD and 13 MEng completed

  • Projects

Projects

– 8 funded projects completed (AcRF, NSTB) – 7 collaborative / industrial projects – various other projects

M M-

  • CMA

CMAC C Associative Associative Memory Memory Hybrid CI Automated Parking System Hybrid CI Automated Parking System Intelligent Well Log Analysis Intelligent Well Log Analysis

– + Input Effort Output Primary Controller Plant Adaptive Controller

X% Y%

Supervisor – + Model

Neuro Neuro-

  • fuzzy Integrated Process Supervision

fuzzy Integrated Process Supervision

slide-4
SLIDE 4

Centre for Computational Intelligence C2i

  • Cognitive Science is

Cognitive Science is

– The study of the mind i.e., cognitive processes and their relation to intelligent behavior, learning, perception, language, emotion, etc. – The key to future intelligent systems, humanized technologies and applications – An important, emerging, multidisciplinary field, fast developing in EU, Japan, USA.

From Cognitive Science to IS From Cognitive Science to IS

slide-5
SLIDE 5

Centre for Computational Intelligence C2i

Research at Research at C C2

2i

i

  • Centre for

Centre for Computational Intelligence Computational Intelligence

  • Adaptive and autonomous systems

Adaptive and autonomous systems

  • Nature

Nature-

  • inspired systems

inspired systems

  • Neuro

Neuro-

  • cognitive architectures

cognitive architectures

  • Decision support systems

Decision support systems

– Synthesize human reasoning capabilities and tolerance to uncertain / incomplete information – Applications in robotics, transportation, HCI, manufacturing, medical, finance, education…

Cognitive Informatics Cognitive Informatics

slide-6
SLIDE 6

Centre for Computational Intelligence C2i

Building Intelligent Systems Building Intelligent Systems

  • From

From

AI AI

to to

CI CI

to to … … MI MI

Fuzzy Computation Evolutionary Computation Neural Computation DNA Computing Quantum Computing

Machine Intelligence Machine Intelligence

Machine Learning Search

adaptation

Logical Reasoning Planning nature- inspired p a r a l l e l s e q u e n t i a l Artificial Intelligence Artificial Intelligence Computational Intelligence Computational Intelligence Probabilistic Reasoning

slide-7
SLIDE 7

Centre for Computational Intelligence C2i

Neuro Neuro-

  • Cognitive Architectures

Cognitive Architectures

  • Convergence of Cognitive

Convergence of Cognitive and Neuro and Neuro-

  • sciences

sciences

  • Study of the human mind

Study of the human mind

  • new science for IS

new science for IS

“ “Example is the way to learn. Example is the way to learn. Example is the only way to learn. Example is the only way to learn.” ”

  • - Albert Einstein

Albert Einstein

  • Science

Science humanized IS humanized IS

– Sensing, diagnosis, guidance – Semantic learning memory – Skill acquisition e.g. driving

slide-8
SLIDE 8

Centre for Computational Intelligence C2i

  • Fuzzy Systems

Fuzzy Systems

– Pros: rules intuitive and easily comprehended, emulates the human decision process, prior expert knowledge can be easily incorporated – Cons: manual design of fuzzy sets and rules, heuristic tuning of system parameters

  • Neural Network

Neural Networks s

– Pros: self-organised learning / modelling, fault tolerance, distributed knowledge – Cons: opaque, no prior knowledge, stability and plasticity dilemma, convergence

Hybrid Fuzzy Neural Systems Hybrid Fuzzy Neural Systems

Fuzzy Neural Networks (FNN)

combine: the advantages of FS the capabilities of NN and other techniques (clustering, tuning, evolutionary…)

Fuzzy Neural Fuzzy Neural Networks Networks (FNN)

(FNN) combine: combine: the advantages of FS the advantages of FS the capabilities of NN the capabilities of NN and other techniques and other techniques (clustering, tuning, (clustering, tuning, evolutionary evolutionary… …) )

slide-9
SLIDE 9

Centre for Computational Intelligence C2i

Example: the GenSoYager FNN Example: the GenSoYager FNN

  • Generic

Generic

: : not not application application-

  • specific

specific

(in control, finance, medical, etc.) (in control, finance, medical, etc.)

  • Self

Self-

  • Organising

Organising

: automated fuzzy set design : automated fuzzy set design and data clustering and data clustering DIC

DIC

  • Yager

Yager

: sound, most accurate : sound, most accurate inference scheme inference scheme Yager

Yager

  • Fuzzy

Fuzzy (rule

(rule-

  • based)

based) : intuitive, understandable : intuitive, understandable approximate reasoning approximate reasoning

  • Neural Network

Neural Network

: rule learning : rule learning RuleMap

RuleMap

+ optimization + optimization GA

GA

slide-10
SLIDE 10

Centre for Computational Intelligence C2i

Automated Fuzzy System Design Automated Fuzzy System Design

Rule 1: IF x1 is C … xn is G, then y is K Rule n: IF x1 is A … xn is I, then y is L

…

Output Input

Fuzzy Rule Base Fuzzy Inference Engine Fuzzifier Defuzzifier

Input x1 Input xn … A B C D E F G H I Output y Defuzzification method: Center of Gravity (CoG) Mean of Maximum (MoM) J K L Inference Scheme and Operators T-norm: MIN, etc T-conorm: MAX, etc

Fuzzy System Construction

1) Generate fuzzy sets to cover the input and output spaces Discrete Incremental Clustering 2) Decide the fuzzy inference scheme and defuzzification method Keller-Yager model 3) Generate fuzzy rules from input-

  • utput pairs -> RuleMap learning

algorithm / rule formulation phase 4) Select the generated rules to form the fuzzy rule base RuleMap / parameter learning phase

slide-11
SLIDE 11

Centre for Computational Intelligence C2i

GenSoYager GenSoYager-

  • FNN System

FNN System

Fuzzification layer Fuzzification layer Antecedent layer Antecedent layer Rule layer Rule layer Consequent layer Consequent layer Defuzzification layer Defuzzification layer

  • 1. Self-organising / Clustering phase
  • 1. Self-organising / Clustering phase
  • 2. RuleMap Rule Formulation phase
  • 2. RuleMap Rule Formulation phase
  • 3. RuleMap Parameter Learning phase
  • 3. RuleMap Parameter Learning phase
slide-12
SLIDE 12

Centre for Computational Intelligence C2i

Evaluation of GenSoYager Evaluation of GenSoYager-

  • FNN

FNN

  • Classification benchmarks

Classification benchmarks

– XOR dilemma – Iris classification – 2-spiral problem – Ionosphere classification

CTE towards Ang Mo Kio PIE towards Changi Towards Upper Serangoon Lane 3 Lane 1 Lane 2 Lane 4 Lane 5

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.2 0.4 0.6 0.8 1

Performs equally or better than previous FNN architectures:

Falcon, Falcon-MLVQ, Falcon-ART/MART, ANFIS, GA classifier, POPFNN(CRI/TVR) …

Performs equally or Performs equally or better than previous better than previous FNN architectures: FNN architectures:

Falcon, Falcon Falcon, Falcon-

  • MLVQ,

MLVQ, Falcon Falcon-

  • ART/MART,

ART/MART, ANFIS, GA classifier, ANFIS, GA classifier, POPFNN(CRI/TVR) POPFNN(CRI/TVR) … …

  • Prediction benchmarks

Prediction benchmarks

– Box-Jenkins gas furnace – Financial data reconstruction – Mackey-Glass time series – Traffic flow modelling

slide-13
SLIDE 13

Centre for Computational Intelligence C2i

Applications Applications

  • Decision Support Systems

Decision Support Systems

  • Classification

Classification

– Forensics/security: handwriting, face identification – Process control: chemical, manufacturing – Medical diagnosis: asthma, cancer, diabetes – Transportation planning, navigation, and control

  • Prediction

Prediction

– Financial forecasting: bank failure, stock et al – Medical prognosis: asthma, surgery – Transportation: traffic flow (macro/micro)

slide-14
SLIDE 14

Centre for Computational Intelligence C2i

Application: Traffic Planning Application: Traffic Planning

  • Dynamic routing of

Dynamic routing of automated taxi fleet automated taxi fleet

– Hybrid ant colony + GA optimization – FPGA realization

JAM

slide-15
SLIDE 15

Centre for Computational Intelligence C2i

Application: Traffic Prediction Application: Traffic Prediction

  • Traffic flow modelling

Traffic flow modelling and prediction and prediction

– Neural and fuzzy-neural techniques

Site

Towards Changi

slide-16
SLIDE 16

Centre for Computational Intelligence C2i

Application: Supervisory Control Application: Supervisory Control

PID Controller Fuzzy Controller M-CMAC Controller Auto- Tuner Model Reference M-CMAC Learning Qualitative Reasoning Intelligent Supervisor Adaptive Control Primary Control Fault Diagnosis

  • Integrated process supervision

Integrated process supervision

– Learning fuzzy rule-based system – Also: car control, image processing

slide-17
SLIDE 17

Centre for Computational Intelligence C2i

Application: Intelligent Vehicles Application: Intelligent Vehicles

Congestions Congestions Congestions Wasted energy Wasted energy Wasted energy Pollution Pollution Pollution Noise Noise Noise Accidents Accidents Accidents

slide-18
SLIDE 18

Centre for Computational Intelligence C2i

Benefits of Intelligent Vehicles Benefits of Intelligent Vehicles

  • Safety:

Safety: reduce road accidents, injuries/fatalities

reduce road accidents, injuries/fatalities

  • Efficiency:

Efficiency: improve usage of the road network

improve usage of the road network

  • Mobility:

Mobility: improve access to goods and services

improve access to goods and services

  • Productivity:

Productivity: improve efficiency and reduce costs

improve efficiency and reduce costs

  • Environmental quality:

Environmental quality: reduce emissions

reduce emissions

  • Flexibility:

Flexibility: adapt to different drivers/passengers

adapt to different drivers/passengers

  • Reliability:

Reliability: improve performance, predict failure

improve performance, predict failure

  • Ease of use:

Ease of use: assist via intelligent interfaces

assist via intelligent interfaces

  • Security:

Security: authenticate user, prevent theft/abuse

authenticate user, prevent theft/abuse

slide-19
SLIDE 19

Centre for Computational Intelligence C2i

Driver Cognitive Modelling Driver Cognitive Modelling

  • Observations

Observations

– Automotive technologies: ad hoc, no integration – Human: superior sensory-motor skills, higher cognitive faculties, integrated and adaptive

  • Motivations

Motivations

– Develop automotive technologies based on a comprehensive study of human behavior – Conceive design tools that will automatically develop and adapt these as and when required – Holistic approach: integrated, multi-modal human-centered intelligent vehicle

slide-20
SLIDE 20

Centre for Computational Intelligence C2i

Driving Skills Inventory Driving Skills Inventory

  • Collision avoidance

Collision avoidance

– Aim: detecting other vehicles and obstacles – Include: rear-end collision avoidance, road departure, lane changing and merging, crossing, parking slot, pedestrian detection

  • Driving assistance

Driving assistance

– Aim: provide location and route information, guidance, and even autonomous control – Include: cruise control, overtaking, reverse and parallel parking, 3-point turns, dock and parking slot identification, etc.

slide-21
SLIDE 21

Centre for Computational Intelligence C2i

Training Methodology Training Methodology

  • Modelling human driving expertise

Modelling human driving expertise

– Naturally expressed as fuzzy rules

  • e.g., IF obstacle ahead and distance is medium and

velocity is high THEN brake is maximum

– Fuzzy rule based system: mapping from perceptual input to control output

  • Method

Method: learning from example : learning from example

– Record input/output data from human driver – Generate a fuzzy rule base (GenSoYager-FNN) – Use the rule base to auto-drive the vehicle

slide-22
SLIDE 22

Centre for Computational Intelligence C2i

System Input/Output System Input/Output

  • Perceptual input parameters

Perceptual input parameters

– Vehicle data: longitudinal / lateral accelerations,

velocity, steering, trajectory curvature, CP indicators

– Lane data: distance to left / right borders, heading

(wrt. the lane), lane curvature, lane delimiters

– Obstacle data: distance from / headway time to front

  • bstacle, relative speed, bearing; blind zones

– Environment data: light, wind, rain, etc.

  • Control output parameters

Control output parameters

– Vehicle controls: steering, acceleration, brake, gear

slide-23
SLIDE 23

Centre for Computational Intelligence C2i

Driving Simulator Driving Simulator

  • Training tool

Training tool

Steering wheel Accelerator pedal Brake pedal Visual display unit

slide-24
SLIDE 24

Centre for Computational Intelligence C2i

Reverse Parking Simulation Reverse Parking Simulation

slide-25
SLIDE 25

Centre for Computational Intelligence C2i

Reverse Parking Simulation Reverse Parking Simulation

slide-26
SLIDE 26

Centre for Computational Intelligence C2i

Microprocessor Microprocessor-

  • Controlled Car

Controlled Car

  • Former RC car with

Former RC car with

– Handyboard – 8 ultrasonic sensors – 2 servo motors – Digital compass – Wireless module

slide-27
SLIDE 27

Centre for Computational Intelligence C2i

Reverse Parking the Toy Car Reverse Parking the Toy Car

ICARCV04-RevPk-RCCar1.MoV

(a) start reversing (b) move forward (c) resume reversing (d) stop (parked)

Centre for Computational Intelligence Autonomous Reverse Parking

slide-28
SLIDE 28

Centre for Computational Intelligence C2i

Reverse Parking the Toy Car Reverse Parking the Toy Car

slide-29
SLIDE 29

Centre for Computational Intelligence C2i

Vision Vision-

  • based Road Driving

based Road Driving

slide-30
SLIDE 30

Centre for Computational Intelligence C2i

Lane Keeping Simulation Lane Keeping Simulation

slide-31
SLIDE 31

Centre for Computational Intelligence C2i

Lane Changing Simulation Lane Changing Simulation

slide-32
SLIDE 32

Centre for Computational Intelligence C2i

Tactical Driving System Tactical Driving System

slide-33
SLIDE 33

Centre for Computational Intelligence C2i

Tactical Driving Simulation Tactical Driving Simulation

slide-34
SLIDE 34

Centre for Computational Intelligence C2i

Conclusion Conclusion

  • Novel GenSoYager Fuzzy Neural Network

Novel GenSoYager Fuzzy Neural Network

  • One

One-

  • pass, fully automated training cycle

pass, fully automated training cycle

  • Validated using various benchmarks

Validated using various benchmarks

  • Performs equally or better than most

Performs equally or better than most classifiers / predictors classifiers / predictors

  • Successfully used to model human driving

Successfully used to model human driving expertise e.g., reverse parking, 3P expertise e.g., reverse parking, 3P-

  • turn

turn

  • Needs: further improve performance,

Needs: further improve performance, interpretability, and the cognitive model interpretability, and the cognitive model

slide-35
SLIDE 35

Centre for Computational Intelligence C2i

Current and Future Work Current and Future Work

  • Realistic simulator

Realistic simulator

– Racer-based engine

  • Real world validation

Real world validation

– Cycab trained to drive

  • Driver behaviour model

Driver behaviour model

– Multi-modal analysis – Eye and gaze tracking