Here is yet another puzzle piece primarily aimed at those doubting that carmakers see huge potential in neuromorphic technology and believing that Mercedes Benz’s implementation of Akida in the Vision EQXX project car’s voice control system was just a one-off experiment (it certainly would have helped if Markus Schäfer had meanwhile followed up with his promised second
In The Loop blog entry on neuromorphic computing).
I happened to come across a weekly podcast series by German national newspaper Frankfurter Allgemeine Zeitung - aka FAZ - called “D:Economy”, featuring current topics relating to digitalisation and technology.
The podcast’s guest in episode 286 (Dec 1, 2023) was Prof. Steven Peters, who - as you may recall - was the founder and Head of AI Research at Mercedes from 2016-2022, overseeing the Vision EQXX concept car project, before returning to the world of academia full-time; he is now Professor for Automotive Engineering at TU Darmstadt. The podcast centred around autonomous driving, and the interviewee was asked for his assessment of the progress made so far.
https://www.faz.net/podcasts/f-a-z-...men-fahren-wirklich-herr-peters-19354415.html
There was one specific passage (from 3:07 min) that I would like to share with you all. After podcast host Alexander Armbruster had introduced his guest, they briefly talked about Steven Peters’ role as MB’s former Manager of AI Research. The latter said that his team had tried to develop and implement ML tools everywhere in the company where they deemed it to be of added value (he listed aerodynamics, user experience, chassis analysis), and when asked about an example of a specific project, he referred to AI-generated routines as an example of how AI features his team had “vorgedacht” found their way into the new E-Class. [The verb used here,
vordenken, derives from the German noun
Vordenker (= forward thinker) and has a very positive pioneering, innovative and visionary connation].
Steven Peters then went on to say the following:
(First a verbatim transcription of the German original, followed by a teamwork translation by DeepL & myself):
Steven Peters: “Wir haben außerdem sehr viele Themen begleitet, die jetzt auch gerade ‘nen sehr sehr großen Hype auslösen, sag’ ich mal - das ist alles, was mit Energieeffizienz und KI zu tun hat. Wir haben das in dem Projekt Vision EQXX damals auch demonstrieren dürfen: Da haben wir die Sprachbedienung erstmalig - nach unserer Kenntnis erstmalig - auf einem neuromorphischen Chip umgesetzt, d.h., der läuft extrem energieeffizient - im Prinzip hat er die gleiche, vor Kunde die gleiche [? etwas unverständlich, evtl. meinte er für den Kunden?] Funktion, es ändert sich gar nichts, nur es läuft eben viel energieeffizienter ab. Jetzt ist die Sprachbedienung keine große Energiesenke in dem Auto, aber es war ein Use Case, an dem man mal zeigen konnte, dass es geht, und unser großes Ziel jetzt - auch in meiner wissenschaftlichen Forschung an der TU Darmstadt - ist, für sicherheitsrelevante Themen, wie jetzt z.B. die Perzeption - die Objekterkennung beim automatisierten Fahren - auf solchen Chips, mit solchen neuronalen Netzen auch eben energieeffizienter zu machen. Und dann sind wir wirklich in einer hochsicherheitsrelevanten, offensichtlich hochsicherheitsrelevanten Anwendung, und das ist noch ‘ne, ‘ne harte Nuss.”
Alexander Armbruster: “Ist Mercedes da eher hinten dran oder vorne mit dabei? Es gibt ja amerikanische Konzerne, die, ähm, zugegeben auch viel mehr Marketing machen, auch bei viel kleineren Schritten sehr große Ankündigungen zum Teil machen, damit aber, wenn man es so vergleicht, ist Waymo, ähm, ist Waymo weit vorne oder fährt‘s mit Daimler auf derselben Höhe ungefähr, oder…?”
Steven Peters: “Also, ich würde tatsächlich sagen, dass Waymo weltweit führend ist, und Waymo benimmt sich, ähm, im positiven Sinne wie ‘ne Universität. Also die, die forschen sehr sehr viel, auch in der Grundlagenforschung, veröffentlichen auch einiges, und das ist wirklich beeindruckend. Und mir wäre jetzt kein vergleichbares Unternehmen bekannt, das in dieser Tiefe und in dieser Ernsthaftigkeit und mit so einem langen Atem dieses Thema erforscht und vorantreibt. Von daher glaub’ ich, sind die zu Recht auf Platz 1, muss man glaub’ ich so sagen. Unsere deutschen sind aber allesamt, äh, jetzt nicht irgendwie abgehängt oder weit dahinter - der Anwendungsfall ist nur ‘n anderer. Und wir sehen das ja - Mercedes war der erste jetzt in Deutschland, BMW ist jetzt gefolgt - mit dem sogenannten Level 3-System; das ist nicht ganz so viel, wie die Robotaxis von Waymo, die in San Francisco jetzt seit wenigen Monaten auch kommerziell im Einsatz sind, aber es ist [sic!] auch weltweit hier die ersten, die an private Kunden solche Fahrzeuge ausliefern, wo ich unter ganz engen, definierten Szenarien als Fahrender wirklich die Hände auch vom Lenkrad nehmen darf und auch die Verantwortung ans Fahrzeug übergebe, d.h. ich darf dann wirklich in diesen Szenarien, wenn das Fahrzeug übernommen hat, z.B. die FAZ lesen.“
Steven Peters: “In addition, we were involved in a lot of topics that are currently generating a lot of hype, I'd say - everything that has to do with energy efficiency and AI. We were also able to demonstrate this in the Vision EQXX project: we implemented voice control on a neuromorphic chip for the first time - to our knowledge for the first time - which means it runs extremely energy-efficiently. In principle it has the same, … [? somewhat incomprehensible in the original, perhaps he meant for the customer?] function, nothing changes at all, it just runs much more energy-efficiently. Now, voice control is not a major energy sink in the car, but it was a use case that showed it works, and our big goal now - also in my scientific research at TU Darmstadt - is to make safety-relevant topics, such as perception - object recognition in automated driving - more energy-efficient on such chips, with such neural networks. And then we will really be in a highly safety-relevant, obviously highly safety-relevant application [more freely translated “we’ll be dealing with…”], and that is still a tough nut to crack.”
[Highly safety-relevant is the literal translation of the adjective hochsicherheitsrelevant, which Steven Peters uses in the German original; I‘d be inclined to use the English translation safety-critical here, but I am not sure whether those two terms would be equivalent in automotive tech speak]
Alexander Armbruster: “Is Mercedes at the back of the pack or in front? There are American companies that, um, admittedly do a lot more marketing, make some very big announcements, even with much smaller steps, but if you compare it like this, is Waymo, um, is Waymo far ahead or is it roughly on the same level with Daimler...?”
Steven Peters: “Well, I would actually say that Waymo is the global leader, and Waymo behaves, um, like a university in a positive sense. They do a lot of research, including basic research, and they also publish a lot, and that's really impressive. And I'm not aware of any comparable company that is researching and advancing this topic in such depth and with such seriousness and staying power, so I think it is fair to say they are deservedly the number one. But our German companies are all, er, not somehow left behind or far behind - the use case is just a different one. And we can see that - Mercedes was the first in Germany, BMW has now followed - with the so-called Level 3 system, which is not quite as advanced as Waymo's robotaxis, which have now been in commercial use for a few months in San Francisco, but it [sic!] is also the first in the world to deliver such vehicles to private customers, in which I as the driver can really take my hands off the wheel under very narrow, defined scenarios and also hand over legal responsibility [liability?] to the vehicle, i.e. in these scenarios, once the vehicle has taken over, I [as the person in the driver’s seat] am actually [legally] permitted to read the FAZ, for example."
So here is my take on our relationship with Mercedes:
While no specific neuromorphic company gets mentioned in the podcast, we know for a fact that it was Akida he is referring to when he talks about the voice control implementation on a neuromorphic chip in the Vision EQXX concept car.
And we just heard it practically from the horse’s mouth (even though Steven Peters is no longer employed by MB) that voice control will not remain the only application to be optimised by neuromorphic technology. Automotive companies have definitely set their eyes even on highly safety-relevant vision applications, too (pun intended). However, it may take a little longer than a lot of us BRN shareholders envisage before neuromorphic chips will be implemented ubiquitously in cars (“… that is still a tough nut to crack.”)
Although we cannot be 100% sure that Mercedes is going to utilise Akida in the future (as opposed to other neuromorphic tech) - unless we get an official announcement - chances are they will stick with the commercially available tried and tested, and the recent reveal by former Brainchip ML intern Vishnu Prateek Kakaraparthi that
@Pom down under had discovered is evidence of continued interest in cooperation between MB and Brainchip until at least August 2023.
Note that the wording is “positioning for
potential project collaboration with Mercedes”, so the way I read it this is no proof of present collaboration, even though other posters have claimed so. To me, it sounds more like MB wants to compare two or more neuromorphic solutions before making a final decision, although that of course begs the question of who could be the potential competition. Last year, Markus Schäfer mentioned both Brainchip and Intel as leading developers in the field in his first
In the Loop blog entry. But how does that align with Loihi not being ready for commercialisation for another few years?
As for the recent CES Rob Telson interview: Yes, he mentions smart cabin features announced by “companies like Mercedes”, but stops short of claiming that it is indeed Brainchip’s technology enabling Mercedes to do what they are promoting. IMO this is no proof either that Mercedes is still a customer.
So while I am convinced that MB (and other carmakers) will implement neuromorphic tech into their future cars and I am optimistic about a continuing collaboration between MB and Brainchip for various reasons (eg as the Mercedes logo continues to be shown on Brainchip’s website under “You’re in good company”), I wouldn’t say it is 100% certain from what we know so far, and I don’t think it is fair to insult people who question the claim made by some that it is or who state they believe the lead times are much longer than what some posters here wish for.
www.linkedin.com
www.tataelxsi.com
thestockexchange.com.au
