ARGUS — The perception brain for self-driving

The live AI

Show ARGUS the road. Watch it think.

Use a sample street, upload your own dashcam clip or photo, or aim your webcam at traffic (or even your screen). The model below is COCO-SSD — a real object detector trained on hundreds of thousands of real photos of streets, cars and people.

Samples Upload image / video Webcam Stop

The AI is asleep

Pick a sample, upload footage, or start your webcam above. ARGUS will wake up, load its neural net, and start reading the road.

Idle — neural net not loaded yet

Drive Brain

STANDBY

Waiting for vision input…

Throttle

Brake

Steer

0°

People0

Vehicles0

Signs / lights0

Tracked0

[boot] Drive Brain online. Awaiting perception feed.

The build

Eight eyes. Full 360° around the car.

A single front camera has blind spots that get people killed. ARGUS is designed around eight cameras — overlapping fields of view so nothing within range is ever unseen. Here's where each one goes.

How it works

From photons to a steering decision in four steps.

Capture

The 8 cameras stream frames. ARGUS stitches them into one continuous view of the world around the car.

Perceive

The neural net scans every frame and labels what it sees — car, person, cyclist, truck, light, sign — with a box and a confidence.

Reason

The Drive Brain ranks each object by how close and how central it is, then computes the risk of the road right now.

Decide

Out comes a command: throttle, brake, or steer — and the reason why, logged so you can always see what it was thinking.

◆ The honest version (read this)

✓ This is real. The detector running above is a genuine neural network trained on real street imagery. It actually sees cars and people — on your device, no internet round-trip after it loads.
✓ It's the real first layer. Every self-driving car starts with perception exactly like this, then fuses it with lidar, radar, GPS and maps.
✓ Already trained on ~330,000 real images. The vision AI learned from the COCO dataset — hundreds of thousands of labeled photos of cars, people, signs and lights. That industrial-scale training is exactly why it recognizes the road the moment this page loads; you can't out-train it on a laptop with a few thousand pictures.
✗ Do not wire this to a real car and drive. A production self-driving system needs sensor fusion, redundancy and billions of validation miles. ARGUS is a perception & decision demo and a simulator brain — not a certified driver.
✓ Where it shines: learning how perception works, prototyping, dash-cam analysis, and driving a car inside a simulator. That's the safe, buildable, genuinely impressive frontier.