Applied neural-network experiment
Neural-network driving simulation
An in-browser driving simulation built without machine-learning libraries. Five distance sensors feed a 5→10→10→4 neural network that decides forward, left, right, and reverse while navigating traffic and road boundaries.
The question
Can a small neural network make a driving decision visible?
The experiment reduces autonomous driving to a bounded system: sense nearby road and traffic conditions, feed those readings into a network, and translate the outputs into four driving controls.
The experiment
The decision path is drawn beside the road.
The browser JavaScript implementation uses five distance sensors and a 5→10→10→4 network built without a machine-learning library. Its outputs control forward, left, right, and reverse while a second canvas visualizes the network activity.
Why it remains
A visible system is easier to reason about.
The prototype remains a compact study in sensing, inference, mutation, and feedback. It makes each layer inspectable instead of hiding the decision behind a packaged model.
Evidence boundary
What this note can prove.
This note records an implemented experiment and the question behind it. It does not turn private source into public evidence or present an experiment as a measured business outcome.
- Private implementation
- The local source was reviewed for the sensor count, network shape, control outputs, and visualization. No public deployment or measured driving result is claimed.