Toyota – Prancing horse.

Toyota comes to its senses (sort of).

• After its inexplicable investment in Uber’s self-driving unit (see here) at a valuation of $7.25bn and its madcap plan to build an entire city in the shadow of Mount Fuji (see here), it has finally done something that I can find some sense in.
• Toyota is investing $400m in Pony.ai at a valuation of $3bn giving it an 11.8% holding in the company.
• In my opinion, Pony.ai is a proper autonomous driving company that has been making pretty good progress and which RFM ranks as global No. 4 in terms of performance.
• On the other hand, Uber’s autonomous driving unit ranks by far the worst in RFM’s analysis, has killed a pedestrian and has engaged in the misappropriation of IP from Waymo that it has been forced to subsequently remove.
• So how Toyota could justify paying twice as much for Uber than it has for Pony.ai remains a mystery.
• One possibility is that the investment in Uber is similar to Microsoft’s investment in OpenAI.
• OpenAI is pursuing artificial general intelligence through massive compute power and consequently will spend almost all the $1bn that Microsoft has invested on Microsoft compute resource.
• Hence Microsoft has essentially purchased a customer and been granted a free option just in case OpenAI’s approach turns out to right.
• Toyota invested $333m in Uber and so the only rational rationale I can come up with is if all of that $333m will be spent on buying Toyota vehicles.
• RFM ranks Pony.ai at global number 4 for two reasons:
  • First, Disengagements: as flawed as this measure is, it is the only one currently available.
  • Here, Pony.ai fares reasonably well coming in 4^th behind Waymo, Cruise and Zoox.
  • Although the disengagement measure can easily be gamed to look much better than it is, it has to date provided a reasonable reflection of reality as supported by RFM’s qualitative research.
  • It is impossible to rank either Yandex or Mobileye by these measures but both of them have impressive demonstrations but have so far declined to disclose any data.
  • Consequently, I am quite optimistic with regard to what they are capable of, but they remain classified as dark horses.
  • Second, Geographic spread: Although Pony.ai remains rigidly geofenced around its test centres, it is testing in Freemont, Beijing and Shanghai.
  • Furthermore, it is doing this with a reasonable amount of code re-use giving me some confidence that its system can be adapted with some work to run in different environments.
  • This is one of the biggest challenges for systems that have a deep learning component to them because the second that something happens that the system has not been trained for, it is thrown into disarray.
  • This is why all true autonomous driving systems are rigidly geofenced to limit the uncertainty that they have to deal with.
  • (Tesla is not geofenced but then I don’t think Autopilot is really autonomous driving either).
• Toyota’s rationale for making this investment makes sense to secure its long position in China and perhaps the USA and it is also at a valuation that I can just about get behind.
• If only it could apply this kind of thinking to what it needs to do with in-vehicle infotainment and the smartphone threat, it would be in a good position to survive the coming decline in unit shipments.
• Of this kind of thinking, I have yet to see any sign, although I am a little bit more hopeful.