self-driving-RC-car

-------------------------------------------

RPI3 mounted on the RC car alongside an USB camera and a power bank.

The raspberry pi will do the capture and some pre processing before sending via TCP/IP socket the bottom half of the flattened frame matrix to a laptop.

The bytes are collected, decoded, and fed into the TensorFlow model where all the computation is done on the GPU.

The resulting steering vector is sent back to the rpi 3 via socket so that it can be used to control the motors using the GPIO pins. The speed at which the car will go forward is controlled simply by changing the freq / duty cycle of a pwm signal.

-------------------------------------------

When it comes to collecting the data set needed to train the model, the RC car is being droven using the left, right arrow keys

on the keyboard while on the laptop frames are being streamed over provinding a first person view.

The sample / label pairs are being stored into an hdf5 file in real time.

--------------------------------------------

Multilayer Perceptron with a unidimensional vector comprised of 38400 of pixel values as input,

one hidden layer consisting of 150 units, 7 units in the output layer which will add up to create a one-hot encoded vector, leading

to 7 possible classes : stopped, slow forward, slow forward steering left, slow forward steering right and the forward, left, right equivalent for high speed.

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
models		models
tensorboard_data		tensorboard_data
utils		utils
README.md		README.md
autopilot.py		autopilot.py
driver.py		driver.py
train_autopilot.py		train_autopilot.py
video_feed.py		video_feed.py

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

models

models

tensorboard_data

tensorboard_data

utils

utils

README.md

README.md

autopilot.py

autopilot.py

driver.py

driver.py

train_autopilot.py

train_autopilot.py

video_feed.py

video_feed.py

Repository files navigation

self-driving-RC-car

-------------------------------------------

RPI3 mounted on the RC car alongside an USB camera and a power bank.

The raspberry pi will do the capture and some pre processing before sending via TCP/IP socket the bottom half of the flattened frame matrix to a laptop.

The bytes are collected, decoded, and fed into the TensorFlow model where all the computation is done on the GPU.

The resulting steering vector is sent back to the rpi 3 via socket so that it can be used to control the motors using the GPIO pins. The speed at which the car will go forward is controlled simply by changing the freq / duty cycle of a pwm signal.

-------------------------------------------

When it comes to collecting the data set needed to train the model, the RC car is being droven using the left, right arrow keys

on the keyboard while on the laptop frames are being streamed over provinding a first person view.

The sample / label pairs are being stored into an hdf5 file in real time.

--------------------------------------------

Multilayer Perceptron with a unidimensional vector comprised of 38400 of pixel values as input,

one hidden layer consisting of 150 units, 7 units in the output layer which will add up to create a one-hot encoded vector, leading

to 7 possible classes : stopped, slow forward, slow forward steering left, slow forward steering right and the forward, left, right equivalent for high speed.

About

Releases

Packages

Languages

catalinHRG/self-driving-RC-car

Folders and files

Latest commit

History

Repository files navigation

self-driving-RC-car

-------------------------------------------

RPI3 mounted on the RC car alongside an USB camera and a power bank.

The raspberry pi will do the capture and some pre processing before sending via TCP/IP socket the bottom half of the flattened frame matrix to a laptop.

The bytes are collected, decoded, and fed into the TensorFlow model where all the computation is done on the GPU.

The resulting steering vector is sent back to the rpi 3 via socket so that it can be used to control the motors using the GPIO pins. The speed at which the car will go forward is controlled simply by changing the freq / duty cycle of a pwm signal.

-------------------------------------------

When it comes to collecting the data set needed to train the model, the RC car is being droven using the left, right arrow keys

on the keyboard while on the laptop frames are being streamed over provinding a first person view.

The sample / label pairs are being stored into an hdf5 file in real time.

--------------------------------------------

Multilayer Perceptron with a unidimensional vector comprised of 38400 of pixel values as input,

one hidden layer consisting of 150 units, 7 units in the output layer which will add up to create a one-hot encoded vector, leading

to 7 possible classes : stopped, slow forward, slow forward steering left, slow forward steering right and the forward, left, right equivalent for high speed.

About

Resources

Stars

Watchers

Forks

Languages