https://www.youtube.com/watch?v=HDSw95Cojp8
🚗 Nearly 1.2 million people die each year due to car accidents - due to careless errors, impaired driving, and reckless behaviour.
⚙️Autonomous Vehicles serve as a potential solution to this problem. Leveraging the high processing speeds of moden computers, such vehicles can make split-second desicions with high accuracies.
💡I decided to implement one myself! As someone who's HUGELY passionate about Artifical Intelligence, I began wondering if I could create a self-driving vehicle using traditional deep learning techniques.
I found Udacity's Autonomous Vehicle Simulator and decided to build a Convolutional Neural Network (CNN) to drive the car automatically around a track! This project uses the philosophy of behavioural cloning - attempting to mimic human behaviour using Artificial Intelligence - to enable self-driving functionality.
After hours of training, I arrived at a model that was actually able to accomplish this. Regardless, the CNN still isn't purpose - I've made some updates to these files as I train better models with lower loss values.
After training is done, the project uses a server-client model using Python anf Flask to actually drive the vehicle. Basically:
- The simulator (client) sends images every second to the API/Server
- The Flaks API/Server activates the model and runs the images through the CNN
- The CNN then outputs a steering angle and sends it back to the server
- The server then sends the steering angle to the client - the car can now drive!
NOTE: For more detail, including mockups and detailed analyses, check out the video above!
All in all, this project was a FANTASTIC accelerator for both my deep learning and programming abilities. Feel free to check it all out!
Regards,
Aditya Dewan