This project focuses on using machine learning to make a remote control toy car to drive on it's own without manual intervention.
- The car is loaded with a RaspberryPi camera, which takes images of what the car is 'seeing' and streams the images to a server (my laptop).
- Depending upon the input image, a classification model (logestic regression/convolutional neural network), running on the laptop, predicts the direction in which the car should move.
- This information is then transmitted from the laptop to the remote control of the car using an Arduino.
-
Rccar.py : This is the main file of the project. The script can be run in the following two modes
- GENERATE_TRAINING_DATA : In this mode, the program stores the incoming images received from RaspberryPi and the corresponding direction which the car makes as a result of keyboard strokes made by the user, and transmits the input direction to Arduino.
- SELF_DRIVE_CAR : The code handles the incoming images from RaspberryPi, makes the prediction and transmits the predicted direction to Arduino.
-
learningModels.ipynb : Deals with data analysis and selecting the classification model.
-
RaspberryPi -> pi_client.py : Sends images from RaspberryPi to the laptop.
-
Arduino -> receivePyCommandAndExecute.ino : Handles the input from the laptop for transmitting the directions to the remote control car.
- Ever since I heard Andrew Ng's lecture on machine learning where he mentions about Alvin, I was fascinated to make an autonomous car myself. So here it is, albeit a much smaller one!
- I found the book: Python Machine Learning by Sabastian Raschka very helpful for data analysis and for learning machine learning techniques in sknn.
- A lot of the project was putting things together from what others have already done in some way or another. This project was greatly inspired from a similar project by Wang Zheng.
- Kenny Fong for helping me with concepts in TCP/IP connections.
- Sneha Rath for helping me a lot with setup and video.
