Dataset for HD maps — comma2k19

less than 1 minute read

Maps are useful. Accurate maps are even more useful. Building accurate maps requires a lot of data. Mapping vehicles are expensive, EONs are not.

“How can cheap EONs ever make accurate maps?”, you might be thinking. Well, let me show you something:

Viewing frusta calculated with sensor fusion algorithm (INS + GNSS + Visual-Odometry). Without map-based corrections (left) and with map-based corrections (right).

On the right side of the image we see perfectly localized camera poses. This was done by using data collected with EONs and grey pandas. But how? Is it magic? It’s not…. it’s just math, and there’s more good news. We want to help you build accurate maps too, here’s some stuff to help you get started:

Updated:

You may also enjoy

Hidden Treasure

1 minute read

We’re giving away free hats with the purchase of a comma two, and that’s great! But what if you already bought a comma two and still want a free hat? We’ve g...

Introducing carrot model

1 minute read

Due to the big success of kale model, we have decided to make more vegetable based models.

End-to-end lateral planning

8 minute read

The goal of the research team at comma is to build a superhuman driving agent. We’ve discussed before how we believe the most effective way to build a superh...

Scaling for 10x User Growth

4 minute read

Weekly openpilot users grew from less than 275 users per week in the second half of 2018 to more than 2,750 users per week at the end of 2020. On some days, ...