A Hybrid Deep CNN-Reinforcement Learning Model for Autonomous Driving / (Record no. 8816)
[ view plain ]
| 000 -LEADER | |
|---|---|
| fixed length control field | 13522nam a22002537a 4500 |
| 008 - FIXED-LENGTH DATA ELEMENTS--GENERAL INFORMATION | |
| fixed length control field | 210112b2018 a|||f mb|| 00| 0 eng d |
| 040 ## - CATALOGING SOURCE | |
| Original cataloging agency | EG-CaNU |
| Transcribing agency | EG-CaNU |
| 041 0# - Language Code | |
| Language code of text | eng |
| Language code of abstract | eng |
| 082 ## - DEWEY DECIMAL CLASSIFICATION NUMBER | |
| Classification number | 610 |
| 100 0# - MAIN ENTRY--PERSONAL NAME | |
| Personal name | Karim Mansour |
| 245 1# - TITLE STATEMENT | |
| Title | A Hybrid Deep CNN-Reinforcement Learning Model for Autonomous Driving / |
| Statement of responsibility, etc. | Karim Mansour |
| 260 ## - PUBLICATION, DISTRIBUTION, ETC. | |
| Date of publication, distribution, etc. | 2018 |
| 300 ## - PHYSICAL DESCRIPTION | |
| Extent | 108 p. |
| Other physical details | ill. |
| Dimensions | 21 cm. |
| 500 ## - GENERAL NOTE | |
| Materials specified | Supervisor: Mohamed A. El-Helw |
| 502 ## - Dissertation Note | |
| Dissertation type | Thesis (M.A.)—Nile University, Egypt, 2018 . |
| 504 ## - Bibliography | |
| Bibliography | "Includes bibliographical references" |
| 505 0# - Contents | |
| Formatted contents note | Contents:<br/>1 CHAPTER 1: INTRODUCTION ................................................................................ 1<br/>1.1 Motivation ............................................................................................................ 1<br/>1.2 Problem Definition ............................................................................................... 1<br/>1.3 Thesis Contributions ............................................................................................ 2<br/>1.4 Autonomous Driving and Artificial Intelligence ................................................. 2<br/>1.5 Organization of Thesis ......................................................................................... 2<br/>2 CHAPTER 2: LITERATURE REVIEW ..................................................................... 4<br/>2.1 Introduction .......................................................................................................... 4<br/>2.2 Autonomous Driving ............................................................................................ 4<br/>2.2.1 5 Levels of Autonomy .................................................................................. 5<br/>2.2.1.1 Level 1: One Control Automated .............................................................. 5<br/>2.2.1.2 Level 2: Two Controls Automated ............................................................ 6<br/>2.2.1.3 Level 3: City to City Automation .............................................................. 7<br/>2.2.1.4 Level 4: Full Autonomous......................................................................... 9<br/>2.2.1.5 Level 5: No Driver .................................................................................. 10<br/>2.2.2 Autonomous Driving Block Diagram ......................................................... 10<br/>2.3 Artificial Intelligence ......................................................................................... 11<br/>2.3.1 Machine Learning ....................................................................................... 11<br/>2.3.2 Deep Learning ............................................................................................. 11<br/>2.3.2.1 Convolutional Neural Networks.............................................................. 13<br/>2.3.2.2 Reinforcement Learning .......................................................................... 14<br/>2.3.3 Recurrent Neural Networks ........................................................................ 14<br/>2.4 Rule-Based Control Systems .............................................................................. 15<br/>2.4.1 PID Controller ............................................................................................. 15<br/>2.4.2 Automotive Rule-Based Systems ............................................................... 16<br/>2.4.3 Advantages and Disadvantages ................................................................... 17<br/>2.5 End-to-End Deep Learning ................................................................................ 17<br/>2.5.1 Advantages and Disadvantages ................................................................... 18<br/>2.5.2 Existing End-to-End Deep Learning Networks .......................................... 18<br/>2.5.2.1 CNN Only Steering: NVIDIA Model ..................................................... 19<br/>2.5.2.2 CNN + LSTM Steering: VALEO Model ................................................ 21<br/>2.5.2.3 Reinforcement Learning Steering: STANFORD Model ......................... 23<br/>2.6 Conclusions ....................................................................................................... 25<br/>3 CHAPTER 3: NOVEL CNN-REINFORCEMENT LEARNING STEERING MODEL<br/>26<br/>ix<br/>3.1 Introduction ........................................................................................................ 26<br/>3.2 Sensors Setup ..................................................................................................... 27<br/>3.2.1 Camera ........................................................................................................ 27<br/>3.2.2 Lidar ............................................................................................................ 28<br/>3.2.3 Steering Angles ........................................................................................... 29<br/>3.3 Deep Network Architecture to Predict Future Steering Angles ......................... 30<br/>3.3.1 Model Architecture ..................................................................................... 30<br/>3.3.1.1 CNN-Only Model Limitations ................................................................ 34<br/>3.3.2 Predicting Travel Path Instead of Steering ................................................. 34<br/>3.4 Adding Vehicle Speed ........................................................................................ 38<br/>3.5 Hybrid Model ..................................................................................................... 40<br/>3.5.1 Free Space Acquisition ............................................................................... 41<br/>3.5.1.1 Cameras ................................................................................................... 41<br/>3.5.1.2 Lidar ........................................................................................................ 41<br/>3.5.2 Real-time Reward Integration ..................................................................... 42<br/>3.6 Conclusion .......................................................................................................... 48<br/>4 CHAPTER 4: EXPERIMENTS AND RESULTS .................................................... 50<br/>4.1 Introduction ........................................................................................................ 50<br/>4.2 Country Road Driving Simulator ....................................................................... 50<br/>4.3 Datasets .............................................................................................................. 54<br/>4.4 Data Augmentation ............................................................................................ 56<br/>4.4.1 Data Processing ........................................................................................... 56<br/>4.4.1.1 Data Flipping ........................................................................................... 56<br/>4.4.1.2 Contrast Modifications ............................................................................ 57<br/>4.4.1.3 Shadow Addition ..................................................................................... 57<br/>4.4.2 Extra Sensors for Data Augmentation ........................................................ 58<br/>4.4.2.1 Angled Cameras ...................................................................................... 58<br/>4.4.2.2 Distanced Cameras .................................................................................. 58<br/>4.5 Results ................................................................................................................ 59<br/>4.5.1 Test Track ................................................................................................... 59<br/>4.5.2 Metrics Used ............................................................................................... 59<br/>4.5.2.1 Mean Square Error .................................................................................. 59<br/>4.5.2.2 Validation Loss ....................................................................................... 60<br/>4.5.2.3 Learning Rate .......................................................................................... 60<br/>4.5.3 NVIDIA Model Results .............................................................................. 60<br/>4.5.4 Predicting Future Steering Angles .............................................................. 62<br/>4.5.4.1 One Second into Future ........................................................................... 62<br/>4.5.4.2 Two Seconds into Future......................................................................... 63<br/>4.5.5 Speed Integration ........................................................................................ 65<br/>x<br/>4.5.6 Free Space Only Results ............................................................................. 67<br/>4.5.7 Autonomous Driving Hybrid Model ........................................................... 69<br/>4.6 Conclusion .......................................................................................................... 74<br/>5 CHAPTER 5: CONCLUSIONS AND FUTURE WORK ........................................ 76<br/>5.1 Conclusions ........................................................................................................ 76<br/>5.2 Avoiding Objects................................................................................................ 77<br/>5.3 Different Vehicles Data ...................................................................................... 77<br/>5.4 Human Brain Comparison .................................................................................. 77<br/>5.5 End-to-End Deep Learning Limitations Overcome ........................................... 78<br/>5.6 Safety and Reliability Considerations ................................................................ 78<br/>6 APPENDIX – DESCRIPTION OF USED LIBRARIES .......................................... 79<br/>6.1 Python................................................................................................................. 79<br/>6.2 TensorFlow......................................................................................................... 79<br/>6.3 Keras................................................................................................................... 80<br/>6.4 Unity ................................................................................................................... 80<br/>7 REFERENCES .......................................................................................................... |
| 520 3# - Abstract | |
| Abstract | Abstract:<br/>Nowadays, to reduce the number of accidents and to increase cars safety levels, all car<br/>manufacturers are racing to reach high level of autonomy to achieve driverless cars as soon as<br/>possible. The conventional ways to achieve the current Advanced Driving Assistance Systems<br/>(ADAS) of controlling the vehicle using traditional control theory proved to be not enough to<br/>handle the unlimited scenarios of autonomous driving in a city environment. It became clear that<br/>the usage of Artificial Intelligence (AI) in terms of deep learning is needed to aid the current ADAS<br/>systems (e.g. Adaptive Cruise Control and Lane Keeping) to reach high level of autonomy using<br/>big data. Currently, the automotive industry is using deep learning excessively in classifying<br/>objects captured by on-board sensors (e.g. cameras), but lately there are many researches being<br/>done to use deep learning to control the steering using Convolutional Neural Networks (CNN) for<br/>raw camera images, or Reinforcement Learning (RL) to learn from mistakes using huge amount<br/>of simulated data. The limitation of the CNN-only solution is that it lacks live-feedback to correct<br/>it-self, and in CNNs using Long-Short-Term-Memory (LSTM) layers to reduce the noise in<br/>steering, they suffer from the need of high computational power on-board. On the other hand, the<br/>RL-only solution is relying heavily on simulations due to the nature of RL of learning from doing<br/>mistakes; therefore, RL cannot be used easily with real-life data. On top of that, both solutions are<br/>not considering the speed of the vehicle as an input but relying only on raw data from the frontcamera,<br/>which leads to a system that only works on constant velocities.<br/>This thesis proposes a novel CNN-Reinforcement hybrid solution where both CNN and<br/>RL are being used at the same time in a supervised mode to provide a live feedback that can be<br/>used in real-life data not only in simulations. It also proposes a major modification in the traditional<br/>CNN architecture to let the neural network predict the path of driving instead of steering angle to<br/>reduce the noise on the output and to avoid using LSTMs which require huge amount of<br/>computational power. Besides that, the architecture has been altered for considering the velocity<br/>of the vehicle to ensure smooth driving at different speeds. The different approaches available in<br/>the research community are reviewed in detail. Experiments and tests are done using computer<br/>simulations on both Hybrid and CNN-only solution and the results are examined in details. Finally,<br/>future work is discussed to enhance the system for better and safer driving. |
| 546 ## - Language Note | |
| Language Note | Text in English, abstracts in English. |
| 650 #4 - Subject | |
| Subject | Informatics-IFM |
| 655 #7 - Index Term-Genre/Form | |
| Source of term | NULIB |
| focus term | Dissertation, Academic |
| 690 ## - Subject | |
| School | Informatics-IFM |
| 942 ## - ADDED ENTRY ELEMENTS (KOHA) | |
| Source of classification or shelving scheme | Dewey Decimal Classification |
| Koha item type | Thesis |
| 650 #4 - Subject | |
| -- | 266 |
| 655 #7 - Index Term-Genre/Form | |
| -- | 187 |
| 690 ## - Subject | |
| -- | 266 |
| Withdrawn status | Lost status | Source of classification or shelving scheme | Damaged status | Not for loan | Home library | Current library | Date acquired | Total Checkouts | Full call number | Date last seen | Price effective from | Koha item type |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dewey Decimal Classification | Not For Loan | Main library | Main library | 01/12/2021 | 610 / K.M.H / 2018 | 01/12/2021 | 01/12/2021 | Thesis |