self_drive

Artificial intelligence automatic driving car based on raspberry pie githubhttps://github.com/Timthony/self_drive

Technological process

Motor control Camera debugging Road data acquisition Build deep learning model, parameter debug Real road simulation of automatic driving Final debugging of parameters

Usage method

1. assemble the raspberry cart hardware.
2. zth_car_control.pyzth_collect_data.py
   Use zth_car_control.py to control the front and rear movement of the car, and cooperate with zth_collect_data.py to operate manually, so that the car can collect data on its own runway. (the process is carried out in raspberry pie).
3. zth_process_img.py
   After data acquisition is completed, zth_process_img.py is used to process the collected data, and some data cleaning work is completed before. (computer execution)
4. zth_train.py
   using neural network model to train data, zth_train.py, get a trained model. (computer execution)
5. zth_drive
   Auto-driving on the original runway can be realized by using zth_drive and trained model in the raspberry dispatch car and loading the model. (raspberry pie execution)

Note: All you need to do is use the code mentioned above. Others are original versions or new modules that are being added.

Matters needing attention

1. ()
   the track needs to be produced by itself, which is very important and determines the quality of data. (I was on the floor, taped with colored tape, and then pasted into the shape of the runway).

the width of the track is about two times that of the body. 3. about fifty thousand or sixty thousand images were collected and thirty thousand or forty thousand were screened out. 4. camera angle problem

Specific production process

4. VNCkeras
5. AWSDpygame
6. wasdVNCwad50000
   0_xxxx,1_xxxx,0w01234
8. .h501234,

1.fine-tuning 2. 3.

NVIDIA end-to-end Model :

Normalization Layer255-0.5-0.50.5GPU

Convolutional Layer5x5kernel2x2strides3x3kernelstrideskernelstridesstridesNVIDIAchosen empirically through a series of experiments that vary layer configurations + NVIDIA3full connectedfc NVIDIAend-to-endfeature extractorcontroller