Self-Driving Simulation

The Deep Learning course in Tongji University (2019 FALL) contains the following contents:

• Machine learning basics: Linear regression and classification, Logistic regression for multi-class Classification problems.

• Artificial Neural Networks: Fully-Connected Neural Networks, Convolutional Neural Networks, Recurrent Neural Networks.

• Applications of Deep Learning: Image Classification, Transfer Learning, Style Transfer, Object Detection, Segmentation, GAN

Assignments include:

• Linear Regression with Numpy

• Logistic Regression with Numpy

• Multi-class Logistic Regression with Numpy

• A 2-layer Fully-Connected Neural Networks with Numpy

• Image Classification with Pytorch

• Style Transfer with VGG19 and Pytorch

As shown in the picture, my final project “Integrated Perception and Decision for Self-Driving Simulation” simulates autonomous driving. The simulator is provided by UdaCity and it contains 3 camera which returns images of the road. I drive the car myself to generate training data, and then I trained a CNN to “drive” the car based on my driving log. The problem of self-driving is quite interesting, but I don’t think it’s proper let a CNN take over since it’s a black box. Maybe we can just treat this CNN as a sensor and do sensor fusion. Personally I hope to work on it in my graduate study. You can find the Github Repo HERE.