hacknjit18-self-driving-car

Creators:

Rahul and Zachary Frank, some assistance from Jacqueline Andrade

Inspiration

• Here.com actually led us to vehicular ideas and it was always a passion to do something with self driving cars for our personal projects and resume.

What it does

• Self Driving AI Simulation that thinks and drives on its own having been trained on environment variables.

How we built it

1. Used Udacity Self Driving Unity Application to create training images and parameters.
2. Built on top of a modelling and driving code based off of research of Nvidia Dave2 Self Driving Car research paper (referenced repository below) and Udacity's testing code.
3. Data Processing on the dataset and transforming the structure to run it off of Google Cloud.
4. Used a super powerful Google Cloud Nvidia Accelerated GPU and trained our model on it.
5. Ran the trained model in autonomous mode on the Self Driving App and analyzed the results.

Challenges we ran into

1. Data preprocessing
2. Instance setting up and environment settings was difficult on Google Cloud.
3. Multiple training runs take up a lot of guesswork and training.
4. Training time took hours.
5. Structure setting and running the application.

Accomplishments that we're proud of

1. We build a self-driving AI that we think was a super interesting application.
2. The time challenges and the stress work was a challenge.
3. Environment settings were difficult.
4. The performance of the trained model was impressive.

What we learned

1. Collaborating and working on a Deep Learning Project.
2. Setting up GPU instances and training on Google Cloud.
3. How self driving cars and CNN's work!

References

1. Nvidia Self Driving Research Paper: https://images.nvidia.com/content/tegra/automotive/images/2016/solutions/pdf/end-to-end-dl-using-px.pdf
2. https://devblogs.nvidia.com/deep-learning-self-driving-cars/
3. Open-Sourced Repository that was based off the NVIDIA Dave2 network topology and mimicked the behaviour into a CNN model: https://github.com/naokishibuya/car-behavioral-cloning
4. Udacity Unity Self Driving Simulator: https://github.com/udacity/self-driving-car-sim

What's next for hacknjit18-self-driving-car

1. Train it for the time that we needed. We wanted to train it more.
2. More environment parameters, cars, speed limits, other challenges for the AI to think.

Built With

• python
• r