Self Driving Car Simulation

Fix Deployment
Change the current structure to Vercel, as requested in feedback , The demo link is just for reference for running the model properly, install it locally.

Also, the project is ML based and will require Pytorch, so hopefully Project will be approved

Final Documentation and Enhancements
These are the one’s I did, took more than 3 hours but the work is finally finished.

Fixed Render WebSocket training and headless evaluation.

Migrated from DQN to transparent tabular SARSA.
Improved native human/agent playback, controls, traffic, and flicker.

Redesigned the frontend with calmer typography, colors, spacing, and CTAs.

The main task was to develop this graphs, it felt impossible but synching is also done !

Added comprehensive architecture, security, algorithm, command, UI, and timeline documentation.

Deployment and Hosting Done
This was a basically a ai ml project still added a frontend for easier acces, if you install the project locally then each feature work perfectly, but for hosting web version just background service without demonstration work.

The first task was to reinvent the UI , figure out small issues like fresh github install will retrain the model from zero meaning all progress in vain. But now it resumes as the attributes are also posted and can be continued from saved checkpoint that are updated when we update the github code. On render it works on persistent history.

Apart from that worked on many issues and finally the project is completed.

Lot of hardwork switching from Streamlit to FASTAPI

Ideation & Initial Code Generation - What began just as school project, didnt thought to be too harsh, even using Ai to assist.
I had this idea to build Self driving car simulation, in which I researched and found we can use reinforcement algoritmns and chose Streamlit for Frotend, then finalised the plan

Tech Stack : - Python
I used Gymnassium - Highway env for Highway Environment, SAARS and DQN for learning and training model. The model is trained using State - Action - Reward system, with control of location thrust, speed lane and more.

There was no novelty just to write it, then I chose streamlit for the visualization of training but turned out to be my biggest mistake, started scaffolding the project, building helper function and script for training, start datbase and Q learning table.

Everything was good, then streamlit code for frontend the challenge I faced was, errors and not able to control environment and controls.

then shifted to native python shell, and optimized gui controls, code shortening, documentations for quick start, control and journals. These are the major parts I did. Almost had 40 + commits unlimited brain storming, decision making still model needs improvement, but I think its NERDEST.

What started as simple prompt journey taught several things, git merge conflicts to logical headache what to do now, at the end everything works fine button everything working and code is structured without AI slop.

If you are reading till now, leave a feedback for me to improve, I know its just a start but small steps count. Whoever is reading this, GOD BLESS YOU.