Level 4 is dead, long live Seeing Machines

Here’s the latest piece of analysis from Colin Barnden, Lead Analyst at Semicast Research on Seeing Machines, Driver Monitoring Systems (DMS) and the auto industry.

“EuroNCAP has moved to 2022*. This is why contracts aren’t being announced, as OEMs and T1s have more time to do evaluations (see Hansen Report). Ironically, the delay takes away the ‘Takata penalty’ hanging over Seeing Machines. Had 2020 stood for camera DMS, pretty much every OEM would have had to go with SmartEye, other Tier 2s or the in-house Japanese Tier 1s. The first half of 2019 is likely to be busy for OEM direct wins, ready for 2022.

Level 4 is dead for mass market vehicles. The trend I see is ‘less autonomy, more DMS’ (L2/3 with DMS). That suggests to me the technically best DMS. The key part of Fovio is the hardware accelerators for real-time vision analysis (and to lower power consumption). ‘Hardware agnostic’ is a trade-off not a free ride. The significance of the 1.3 bn kms RNS in the summer is also now clear. Artificial Intelligence/Machine Learning is all about quantity of data. I see Seeing Machines even put live updates of the total on their website. This is smart.

OEMs are acutely aware of regulatory and political threats. Dieselgate was a disaster and emissions in general has been handled poorly. Now the political threat is number of road deaths (hence Vision Zero) and that issue is also being dumped on OEMs’ doormats. Waymo and robo-taxis are an existential threat, OEMs have got to find a way to reduce fatalities fast and win political points. They won’t mess up twice and DMS is the obvious way to proceed. Again that suggests technical excellence over anything. If they are smart, OEMs will ‘front run’ the politics and put DMS into everything as fast as they possibly can. There could be a huge ramp from 2023-2025. Again, a fast ramp up supports longer evaluation times and careful decisions for T1s and T2s.

That’s as far as market analysis can go. What matters now is the actual decisions OEMs make. My role is to make an argument but it is up to everyone to make their own individual decisions about how they think things will play out. No one has a crystal ball.”

*’Europe on the Move’ announced Advanced Distraction Recognition (camera-based DMS) from September 2023. EuroNCAP 5* requirements are looking like they will move to demand camera DMS about a year before.

Chris Menon holds Seeing Machines stock.

Semicast Research on Seeing Machines

I recently wrote to Colin Barnden, Lead Analyst at Semicast Research, asking him about a tweet he sent on October 7th, in which he wrote:

Consider…Denso abandons DN-DSM for trucks to license manufacture of best-in-class Guardian from @seeingmachines. Exits relationship with FotoNation & signs non-compete agreement with SM for DMS, this clears Toyota to appoint Denso/SM to supply Fovio DMS for all cars & pickups.”

I’d assumed he was referring to auto and Toyota primarily but he wished to clarify at length this and other matters on which he disagrees with what I’ve written.

Therefore, in the interests of making Safestocks a forum for genuine debate to the benefit of all investors, I’ve included his comments in full.

Colin Barnden

“You have a number of different issues mixed up, both here and recently. I’ll have a go at unpicking some of it.

Firstly, the tweet was about Guardian and Denso, not auto and Toyota. SM have decided to abandon anything to do with contract manufacture and are heading for an IP-only business model. The problem with Guardian is the price/volume vortex – which is where Tesla are stuck. There is nothing wrong with Gen2 per se and SM have sensibly decided they cannot pour any more money or resources into manufacture and distribution. Dumb would have been to raise say £250M and try to become a global aftermarket company (suicide…and as an investor you’d have been wiped out). Smart is to let someone else with the necessary expertise make Gen2 and SM use the data and take the monthly SaaS revenues. No point reinventing the wheel.

There’s realistically three companies I see who are established aftermarket truck equipment suppliers that Guardian would make sense to go to: Bosch, Conti and Denso. Of those, Denso has already developed a competing product (DN-DSM) so it follows they see the value and are probably most interested. Take DN-DSM and FN out of the picture and Guardian is good to go. The point being if you make those relationship changes, it might open the door to an automotive agreement with Toyota too. That is just speculation (hence the word “Consider…”) and a Toyota deal is at the end of this long chain of events. However Guardian could bring Denso closer to SM (I believe Bosch is working with OEM2 and Conti with OEM3) so there are strategic benefits for both SM and Denso to look into this. Time will tell of course.

Where Sanjay at Panmure gets “no value” for Guardian from is a complete mystery to me. I believe the CAT deal was worth about US$17.5M [CAT vehicles in operation ~3 million units]. For truck/coach/bus it is probably more like 500 million vehicles in operation, so what is the value of a licensing deal for Guardian to say Denso to access that market? US$25-50M??? Maybe much, much, more could be on the table, so it looks to me like a significant cash injection is on the way in the short term. You can’t forecast it and it is a binary outcome (either yes or no) so can’t put in a financial report. SM is an IP company, there are other players better suited to Guardian manufacture and marketing for aftermarket and all that has happened in recent weeks is a change in the go-to-market strategy.

Guardian is also the data gathering platform. In August it was over 1.3 billion kms of naturalistic driving data. The rate of gathering is unknown but it is clearly more than 100 million kms a month and in a world of AI and ML the company with the most data wins. Always. That data is captured in the form of video clips which are sent via 3G/4G to the R&D team (Mike Lenne, Tim Edwards et al.) who are doing analysis and further development of the algorithms. If you read the placing documents from last year you will see one of the areas that SM was going to spend a lot of the money on was advanced scientific research equipment. What you have then is the video clips showing areas for new research (edge cases), the scientific equipment to understand what happens to humans in those circumstances and the results fed back into improvements in the algorithms to improve fatigue and distraction detection further.

It would be great to think that you can skip the scientific research bit and all that is needed is ML and enough compute power on a GPU to do everything perfectly (a la the Nvidia pitch) but that just leaves a solution consuming huge amounts of power and kicking out lots of heat (which is roughly where Affectiva are). There is no short cut to the research, science and sheer hard graft to understand human fatigue and distraction and get the best performance/power consumption trade-off. This work is hard, laborious and necessary. Guardian is the feedback path from the test bed (humans operating in real world trucks) to the R&D/science lab (Mike, Tim and team). To be frank, this is some of the smartest joined-up product development I have ever seen, all pulled together under the leadership of Paul Angelatos and Ken. The staff totally deserve their share awards in my view, the strategic thinking and foresight is extraordinary.

On to auto…from a tech perspective, automotive is not a contest. You just can’t compete with an FPGA solution with either an MCU or GPU for DMS. For vision processing you need hardware acceleration to do it in real time, and Fovio can do that. Add in the 1.3 billion kms of data from Guardian and you have a platform that is untouchable. And it is and OEMS know that. DMS is a crowded market and competition for SM extends well beyond SE (Aisin Seiki, Can Controls, Clarion, Eyesight, FotoNation, Idemia, Mitsubishi, Omron, Panasonic AIS, Pioneer). With Fovio FPGA and 1.3 bn kms of data, the competitive position really comes down to SM vs. all others. It is not yet clear if OEMs want high performance or low price but that will become known over the next 6-9 months as the DWs are announced following the RFQs in March/April. Due process takes time. Also I completely disagree with comments that SM have DWs that they have not announced in the form of an RNS. I do however believe there may be OEMs that have given a verbal nomination and the legal agreements are being worked on. That takes time to work through, there are many decisions to be made for a new vehicle model and it is a complex process which really does take months.

What I have learnt from my experience as an analyst is the company making the most progress is always the one making the least noise (why tell anyone ANYTHING if you are ahead?). Mobileye is probably the best known example for you, but others where this was true are Qualcomm, Broadcom, Marvell. As an analyst researching those companies in the early days, you just hit a brick wall. Apple is the same -ARM staffers sometimes refer to them as ‘the folks at Cupertino’ –  so too Xilinx and Nvidia (guess which has more silicon revenues in production vehicles). SM realistically only need to work with 15-20 OEMs for worldwide coverage, and probably the top ten is enough (everyone else will follow the decisions made by the leaders). So really the lack of information flow coming out of SM I see as evidence of their competitive leadership, which is blatantly obvious when put alongside using FPGA for the silicon solution and the 1.3 bn kms RNS. This is the perspective that almost 25 years of experience gets you.

So my view is unchanged…a super smart company adapting its strategy (Guardian) to husband its financial resources and looking very well placed to take a leadership position in automotive when DMS takes off around 2021. Can’t see the BoD going for a takeover before then, unless someone like Apple or Waymo comes knocking with an unbelievable offer… which is a scenario we have talked about previously.”

Chris Menon holds stock in Seeing Machines.

Seeing Machines is worth US$10, even £10, but not 10p

At the Automated Vehicles Symposium (AVS) held in San Francisco last week, one presentation made was by the National Transportation Safety Board (NTSB), about the first Tesla crash involving Autopilot. The NTSB said that “steering wheel torque is a poor surrogate measure” for driver attention. In a tweet highlighting the presentation, Colin Barnden, Lead Analyst at Semicast Research commented: “This only really leaves camera-based DMS to fulfil driver engagement function.”. In a subsequent tweet Colin also identified a possible scenario where Waymo buys Seeing Machines, maybe even in a 12-18 month timeframe, for US$10 billion.

Here’s Colin’s reply in full to my asking about his thinking behind these tweets and the jaw dropping valuation.

Colin Barnden

The NTSB presentation at AVS. That’s a game changer. If you are a transport executive and you value your freedom, you don’t ignore NTSB recommendations. This even applies to anyone with the first name Elon too.

Level 3 is starting to gain traction so Waymo are looking like they have called the handover problem incorrectly and L3 is possible after all. Time will tell on this. L2/L3 is where the volume will be in my view, at least for the next decade.

Robo-taxis may get investor and press attention, but the volume will be in the mass market. Seeing Machines is the classic ‘pick and shovel’ play, the tech can go almost anywhere in transport applications that humans and machines interact. It certainly isn’t obsolete.

Price… who knows? Could be higher, depends how desperate the bidding war gets (see Sky as a good example). Remember what I wrote to you last week “I can see ten bucks a share persuading the Board to sell up soon, or even ten pounds, but not pennies. That would be stupid, and they (the Board of Directors) aren’t”. [This refers to us discussing privately the likelihood of Seeing Machines’ management accepting a low-ball bid in the next few months].

The current market cap simply reflects that the market is clueless to what SM has achieved. The company isn’t clueless, the executive management are whip smart. The market is coming to them (and Smart Eye too) it just needs patience. Maybe even as little as 12-18 months.”

You can follow Colin at @semicast_res

You can follow me, Chris Menon, at @Penforjustice

The author holds shares in Seeing Machines

Battle of the Titans?

In response to my latest blog post, Colin Barnden, Lead Analyst at Semicast Research, wrote to me explaining why he thinks Fovio is of strategic importance to both Apple and Alphabet. Indeed, his analysis reinforces my feeling that the two may soon battle it out to acquire Seeing Machines — regardless of any initial low-ball bid in the 25p-30p range from another party that kicks off a bidding war. 

I’ve reproduced his comments in full below so you get the benefit of his insights:

Colin Barnden

“Do you hear the thud? That’s the sound of the penny dropping in Silicon Valley that autonomous driving is not easy peasy after all. Witness no robo taxi development stories since March, after the Uber crash. As you report today, witness also the speed of conventional OEMs starting to adopt camera-based DMS. This is a technology which I have repeatedly been told by people at Silicon Valley based tech companies is “at best an interim solution and at worst already obsolete”. Elon might think something like that, but a steady stream of auto OEMs seem to want to work with Seeing Machines anyhow. My view is that every auto OEM will have announced their plans for camera-based DMS by the end of this year, with most implementing the technology for production from 2021 to 2023.

Apple have been trying to get into automotive for several years. Project Titan never really got off the ground, but CarPlay has been well received. For Alphabet, they have to hedge their position that Level 3 is redundant and have already moved to Level 4. There are many articles discussing this, here is one: https://www.wired.com/2017/01/human-problem-blocking-path-self-driving-cars/

I totally disagree with that view and Level 3 is very much possible, but it needs advanced DMS and sufficient human factors research to understand what humans do in the seconds and minutes following handing over driving to machine intelligence. Seeing Machines are underway with this work, led by Mike Lenné and the CAN Drive project. As Tesla have found – and Cadillac proven – you cannot even do Level 2 safely without camera DMS and the recent EC legislation calls for mandatory advanced distraction DMS (camera-based) even for Level 0. Other regions will follow in my view.

Both Apple and Alphabet need it, ideally so the other doesn’t have the technology. Whoever wins gets a minimum 3-5 year lead on the other (automotive qualification alone is 3 years – and Fovio is automotive qualified). Remember Zuck bought WhatsApp for $19 billion so Facebook had a potential rival under control. Conventional metrics for valuing a company won’t apply, this would be a straight Battle of the Titans. I’ll be watching.”

The writer holds stock in Seeing Machines.

EU mandates DMS for 2020

As part of a move to make European roads safer the EU today recommended making driver monitoring systems mandatory for new cars. It’s a momentous decision that is great news for Seeing Machines, causing its share price to shoot up today and Cenkos to upgrade its price target from 10p to 16p.

I asked Colin Barnden Lead Analyst at Semicast Research for his reaction and this is it:

Colin Barnden: So the paint is still wet from the announcements today, but I can draw some basic conclusions.

The EC has today announced the WHEN for a series of automotive systems to become mandatory, a list which includes distracted & drowsiness driver monitoring systems (DMS). My reading of the timeline is as follows:

  • 1 September 2020 for Type Approval (the certification process for new models, facelifts, major changes etc.).
  • 1 September 2022 for all other new cars, irrespective of Type Approval.

That is a two year phase-in period, and from today to the beginning of September 2022 gives a little over four years. That is a sensible timeframe over which to introduce a new technology like DMS.

The EC appears to have left the WHAT of a DMS to be defined by Euro NCAP and I expect the specifics of DMS requirements to follow in short order. This doesn’t guarantee that camera-based DMS will be mandated, nor specified by Euro NCAP. However looking at some of the other systems on the list (AEB, LKAS, ISA, EDR) these are all electronic systems so it would not follow that a mechanical (and very poor) DMS would be specified, even more so for a primary safety feature. Also the entire focus of this regulatory announcement is road safety. I regard AEB, ISA and LKAS as systems needed to compensate for distracted and drowsy drivers, so the obvious way to improve road safety is not so much to correct speed and steering errors, but to keep hands on the wheel, eyes on the road and minds on the task [of driving]. The very best DMS is the obvious way to do that and that points to a camera-based system, from the likes of Seeing Machines for example.

There are two excellent articles I would suggest as backgrounders for your readers for DMS:

http://www.thedrive.com/tech/20843/elon-musk-reportedly-rejected-driver-monitoring-for-tesla-autopilot-but-why

https://www.eetimes.com/document.asp?doc_id=1333236

If I were to also speculate on the WHERE; my view is that Japan will go next for mandatory DMS, probably followed by South Korea. Also, OEMs in the US are on a voluntary agreement (not mandated by NHTSA) to introduce AEB by 2022. I would speculate that they would be highly likely to add DMS on the same timeline.

The key is now what Euro NCAP announce.

Chris Menon: What exactly is the difference between ‘Type Approval’ cars and other new cars?

Colin Barnden: Okay. So, as an example, take the VW T-ROC. It was just launched and would have gone through Type Approval probably late last year. Type Approval is a bit like a new car MOT, someone looks over it and says: “Yes it has seat belts, yes, it has airbags, yes, it has brakes etc.” If it meets all the legislation in place at the time it gets… ‘Type Approval’. Provided no changes are made to the car it maintains Type Approval and no more need be done. Sometimes this can last for four or five years until the model is refreshed. However, from 1 September 2022, new T-ROCs would need to have DMS added, not for Type Approval (which it already has) but to be legal for sale.

Now take a next generation Golf going for Type Approval on 2 September 2020. The list will have another entry “yes, it has DMS” and you get Type Approval. If it doesn’t have DMS, it doesn’t get Type Approval and cannot be sold in the EU.

So on 1 September 2020, only cars going through Type Approval will have to have DMS (or a bit before most likely). By 1 September 2022, it will be in 100% of new cars sold in the EU. This is how the phase-in is managed.

Seeing Machines compared to Mobileye

I recently asked Colin Barnden, Lead Analyst at Semicast Research for his views on Seeing Machines. I’ve reproduced my original questions and his reply in full, as his insights are worthy of a wider audience and deserve to be accurately reported.

Chris Menon: I’m very keen to find out what you think might be the likely valuation of Seeing Machines in the event of a takeover, if you’d care to speculate. Can it be likened to Mobileye in terms of its dominance of DMS? I’m also eager to know if you think there is much real competition? From what I hear Smarteye is a very distant second and its technology is in no way of comparable quality or reliability.

Colin Barnden: “I’m a market analyst not a financial analyst so the issues of valuation are out of my areas of expertise. That said, I don’t think there is a single financial analyst who could accurately value Seeing Machines (SM) as the company is active in so many markets and at so many points in the supply chain. SM also seem to be creating markets as they go along, which is highly cash intensive and has a long “time-to-money”. However get the strategy right and the rewards can be extraordinary. See Google, Facebook and Netflix as examples.

Mobileye is probably a good comparison to SM. Yes there is plenty of serious competition in DMS but what I see tends to happen in IP markets is that one company dominates and then everyone else is competing for what’s left. For example Mobileye has something like 65% of the automotive front camera market, with Xilinx the clear number 2. Which Tier 2 becomes number 1 for DMS depends largely on whether price or features matters most to OEMs.

I suspect it will be features…here is a document I have been reading that I believe pre-announces changes to vehicle legislation [for automotive] for the EU, to be made on May 16: https://www.governmenteuropa.eu/important-year-vehicle-safety-europe/84888/

My reading of it is that DMS becomes mandatory for all cars in Europe from 2020 and with a focus on both drowsy driving and distraction. That suggests camera-based DMS eye-gaze tracking for distraction and PERCLOS (PERcentage CLOSure) eyelid measurement for drowsiness. This is really complex to do well and not many Tier 2s can. The mention of an event data recorder also suggests a Tier 1 might go for a more complex DMS in order to save cost on the DMS/EDR combination. I also read into the announcement that alcohol impairment detection is likely to be a future feature for DMS.

I don’t cover trucks but the legislation there tends to front-run that for automotive by a few years. I really would not be surprised if DMS was made mandatory in Europe for all trucks and buses too, and to my knowledge SM is in a party of one for aftermarket fleet systems (with Guardian).

I’ll be watching on May 16  to see what the EU formally announces. If they mandate everything listed in that article, that would be a step change in road safety. In my view DMS will be the story of the 2020s, with autonomous driving not likely in any meaningful volume until the 2030s.”