A 2020 Theme: Quality

3 minute read

The comma two was a huge success, comma is now a profitable company!

This was a big milestone for us. Of course, the work is not close to over. Our mission is:

To solve self driving cars while delivering shippable intermediaries

What this means in practice is that we aim to produce a piece of software capable of driving better than a human. This is a long term goal, so it’s important to us to deliver value along the way. A piece of software that nicely complements a human can add quite a lot of value, as our users will attest to.

The self driving car levels do not capture the full picture. They don’t say anything about the actual quality of the system, but more about liability. We plan to be Level 2 for a long long time, even when the software can handle all of the driving task.

How we measure value

Firstly, we need to build a stable system. Humans have a MTBF of like 500,000 hours. Our goal with openpilot 1.0 is 1,000 hours, and we should reach this goal this year. This is the job of the openpilot team, in conjunction with the hardware team.

Secondly, we need to build a system that rarely makes mistakes. Humans make mistakes about every 100,000 miles. Our goal with openpilot 1.0 is 100 miles, and we should also reach this goal this year. This is the job of the research team. Note that this number can’t really be compared to the disengagement numbers of L4 companies, as openpilot tries to work everywhere you engage it, not only in a premapped area.

No Discontinuities

Last year, we shipped a model capable of steering the car end to end (aka laneless). We are still using lanes in some scenarios while we build out our localizer, but check out this video. We can keep the human path through intersections, railroad crossings, and the like.

This year, we should ship a model capable of doing the gas and the brakes end to end (aka carless). Just as it was easy to add lane changes once we had end to end lateral, fingers crossed that it should be easy to add traffic lights and stop signs once we have end to end longitudinal.

After both lateral and longitudinal are joint and end to end, there should be no more discontinuities on the gradient of progress to superhuman self driving.

We build software

In 2020, we suspect many self driving car companies will start to falter. They have taken way too much money and hilariously overpromised, and by the looks of the Cruise Origin, seem to really lack direction. The world doesn’t need more bus looking things, it needs better software.

We aren’t looking to build a business and artificially pump it for an acquisition. We aren’t looking to hire marketing and biz dev people to hype up half baked software. We built a business just to fund the software development effort.

We are looking to build the driver assistance system you won’t be able to avoid thinking about because it’s just that good, like what Linux did for operating systems. And with it being 100% open source, anyone can take it and build more value on top of it, without worrying about how we might try to rent seek.

Join us

If you are motivated by producing great open source software that delights thousands when it gets better with every release, consider a job at comma.

And if you just want to watch fireworks, follow us on twitter.

Updated: