Summary: This short post explains why I have not been posting, talks a bit about the upcoming Stuttgart symposium, and discusses the ecosystem
I have not been blogging much lately (though I have been learning lots of new stuff). Here’s my excuse:
It’s that Stuttgart time of the year again: I have been going to the Stuttgart Autonomous Vehicles Test & Development Symposium for a few years now – it seems like a good place to learn about this topic (here are my reports from the previous three installments).
I am going this year as well. But also, perhaps somewhat hastily, we (Foretellix) decided to actually demo our stuff in the associated expo, so the last few weeks were somewhat hectic. That demo is still not nearly where we would like it to be, but I think it conveys the general vision of where we are going.
Six of us will be there (June 4..7), so hopefully we’ll be able to man the booth, attend the symposium (some pretty interesting presentations on the agenda) and talk to people. If you want to chat, drop us a line (info@foretellix.com) or just come by our booth (8009).
I am especially interested in knowing what we are doing wrong (or better yet, what problems we are not solving which we perhaps should).
That ecosystem: The ecosystem around AV verification seems even more complex (and interesting) than I assumed.
We have been talking to a quite a few execution platform providers (simulator providers, HIL providers, people who plan automated test tracks etc.), mainly to explore interface issues. Some of those simulator providers do pretty impressive stuff (and, perhaps unsurprisingly, they each explained why they are significantly better than all others).
Anyway, for most aspects there does not seem to be much interface commonality (so we plan to work with them on that). One exception is the topology aspect, where there seems to be a growing consensus to use OpenDrive: Nobody really loves it, but I guess everybody finds it useful enough.
For that initial demo, BTW, we are using Carla, an open-source simulator (because it is open-source, and to emphasize that we are execution-platform-neutral).
That ecosystem includes the obvious suspects like OEMs, tier 1 and full-AV-stack suppliers, but there are lots of others.
For instance, some of the most interesting discussions we had were with certification/regulation bodies and related consulting companies (we’ll be talking again to some of them in Stuttgart). And I was pretty impressed by how open most of them were.
Expect a post-Stuttgart post: This is going to be an interesting conference, and I promise to blog soon about what I’ll see there.
Thank you for another interesting post. I will patiently wait for your demo.
A thought, following a recent TV item claiming that Mobileye is teaching “street smarts” into its system: once there is an environment with multiple independently developed and certified AVs, you may get some interesting interactions between, say, Waymo vehicles vs. Teslas vs. others. For fun, imagine one specie aggressively merging into traffic, another batch honking furious alerts (but not after 11 PM), and yet another “nerdy” strain playing it so safe they never get anywhere. Technically it’ll be an environment that receives step-changes via uncoordinated SW upgrades by key players, and cross-training in the field may create interesting resonance-like phenomena. How do you verify coherency (or sufficient noise for it to make no difference)?
Hi Nir
That’s a good question I did not think much about until now.
We did assume the need for verifying AV-to-AV interference of various kinds (e.g. many will have 900nm Lidars and perhaps similar Radars, and so cross-vehicle sensing mistakes should be verified). I also discussed verifying human-AV common assumptions in [1].
But I agree that the need to verify the interactions of AV fleets upon new SW release is an interesting, and non-trivial, challenge.
[1] https://blog.foretellix.com/2017/01/26/verifying-interactions-between-avs-and-people/