Putting it all together
If you're still with me, let's try to piece all of this autonomy information together, and let's see what Tesla's chances are in this industry. To do so, let's look at how well Tesla is positioned in each of the things important to creating an FSD system.
- Perception. Vision is the most important element here. You can take a shortcut by using LIDAR, but you're going to need a world class computer vision system eventually to get to full autonomy. HD Maps in theory is a good alternative to vision for localization, but in practice it is extremely hard to make it work because of the scale required to make it work worldwide, and because they need to be kept up-to-date. Only Mobileye's mapping solution seems like it could work, but why not just do localization through vision. Tesla is one of a few companies entirely focused on solving vision, and among those companies Tesla is the clear leader.
- Planning. I haven't talked much about approaches that companies have to planning, because there isn't a lot of information available. Tesla showed a number of impressive examples during their Autonomy Investor Day of how they use things like shadow mode to test and refine new features, but I have not been able to find anything about how others' approaches may differ.
- Iteration. Real world data is everything here, and nobody has a fleet that can be used for real world testing and data gathering anywhere close to the size of Tesla's. As a matter of fact, all of Tesla's competitors combined don't even have a fleet that's in the same ballpark as Tesla's. Furthermore, Tesla's production (and therefore Tesla's fleet size) is increasing exponentially, so this is pretty much Game, Set, and Match. And I haven't even talked about things such as Operation Vacation, that once completed will supposedly automate Tesla's entire training & iteration process, so that Tesla's Autopilot team can go on a vacation, hence the name.
- Hardware. As I've talked about, Tesla's autonomy chip is at least one to two years ahead of the competition's chips from Mobileye and Nvidia.
- Regulatory Approval. Nobody, and I mean absolutely nobody, has a clear path to proving their system saves lives, except for Tesla. Mobileye's RSS seems like a moonshot that's unlikely to convince regulators. Tesla is the only player that will be able to gather enough data to prove beyond any reasonable doubt that their system saves lives.
Taking this all in, I think the question we have to ask ourselves is not who's going to be the first to solve autonomy, that will be Tesla for sure. The question we need to ask ourselves is whether anybody else will be able to solve autonomy before the point of no return beyond which Tesla's monopoly is inevitable. I think the answer to that second question is quite possibly "no".
Waymo does not stand a chance. Their strategy is flawed in that they rely on LIDAR and an effectively impossible to scale HD Maps system. Their fleet is also too small. Other companies using LIDAR are in even worse shape than Waymo. All the same flaws, but years behind.
The only company using LIDAR that stands a chance is Mobileye. It may be better to use vision for localization, but their HD Maps solution works at a large scale. Moreover, they are developing a vision only system in tandem to their LIDAR system. Publicly they say this is to build two complimentary systems that work together, but it could be that internally they feel like they should hedge against LIDAR being a dead-end, or that they already see the writing on the wall and have already shifted most of their focus to the vision only system. Mobileye also makes their own hardware chip that's about as close to Tesla's as can be (one to two years behind), but the crucial ingredient that Mobileye is missing is large scale real world testing and data gathering. They do not have a fleet like Tesla's, and without that regulatory approval could be a problem, and most worryingly it will take them a very long time to build a vision system powerful enough to support full autonomy. In the words of their own CEO: "I don't believe there's technology that can meet those probabilities of failure. Why do I think this? I know what the probabilities are in ADAS, and we are the leading company in providing these.".
So that leaves the companies not pursuing LIDAR. Can any of them hope to compete with Tesla? Comma.ai to be frank does not seem to have their stuff together, so I don't think they will. Pronto.ai? Maybe, but are they really going to find enough trucking companies willing to pay their 'deeply discounted price' of $4,999
per truck to build a meaningful fleet? That seems unlikely.
Therefore, in my eyes it appears very possible that Tesla will monopolize autonomy.
