Deciding what a LiDAR has to be able to see

Blickfeld was building LiDAR around its own MEMS mirror technology — its beam steering, integrated with off-the-shelf components everywhere else, which is how the sensors ended up in a form factor small enough to be genuinely attractive to a carmaker. The company wanted into automotive, and automotive is not a market you enter by having good hardware.

So they built a small team to go at it: a physicist, a software and systems engineer, a VP of business development, and me. Four people, reporting directly to the COO, Terje Noevig. I was there twenty hours a week alongside the master’s.

My job sat in the gap between business development and systems engineering, which is an uncomfortable and very educational place to sit.

The question nobody could answer directly

A Tier 1 customer does not ask “how good is your LiDAR”. They ask whether it is good enough for what they are building — and that turns out to be a question the sensor company has to answer on their behalf, because the customer will not hand over their own requirements.

So the work was to derive them. Two sources, decomposed separately:

• Regulation and NCAP. Euro NCAP and the regulatory framework state what an ADAS function must achieve — this scenario, that closing speed, that outcome. None of it mentions LiDAR. Getting from a required braking outcome to a required detection range, and from there to reflectivity, resolution, point density and field of view, is a chain somebody has to build.
• The Level 4 driving task itself. Not a regulated function but the whole job: everything an autonomous vehicle must perceive to drive unsupervised, and what each of those perceptions demands of a sensor.

Out of both came stakeholder requirements for the sensor — an argument, in the customer’s own terms, about what the hardware had to do and why.

Why it turned out to matter

I did not know it at the time, but this is the first time I did the thing that later became my first delivery at Plus: take a body of regulation that says nothing about your product, and turn it into traceable requirements on it. At Blickfeld the corpus was NCAP and the output was a sensor specification. At Plus the corpus was UN, EU and German road vehicle law and the output was stakeholder requirements for an autonomous driving system. The technique is the same one, and I learned it here, part-time, at a startup.

There is a second thread too. Blickfeld was about specifying a sensor. Argo AI, a year later, was about verifying what perception does with one. Plus is about validating the system all of it adds up to. I did not plan that order, but it is difficult to imagine a better one.

The trade fair

The other half of the education was commercial. The role came with a lot of exposure to automotive Tier 1 customers, and I was part of the small team on the booth when we represented the company at the IAA in Munich — the first year the show had moved there from Frankfurt.

Standing at a stand explaining your technology to people who have no obligation to be impressed is a very fast way to learn which parts of your argument are load-bearing. I have been better at explaining engineering to non-engineers ever since, and I trace it directly to those days.