But a guy here keeps telling us Tesla's Autopilot is already great.
I
absolutely don't believe in the "all or nothing" approach because
nothing ever changes if we insist on waiting for perfection; I believe in
iterative development as a way to improve on existing defaults (ex. human error in driving). The current iteration of Autopilot IS fantastic:
1. The progress has been good, especially for an 11-year-old technology
2. It's the best system available for consumers on the market as it stands today
3. It's getting better every day! As with AI, "today will be the worst day the technology will ever be"
However:
1. It is not perfect
2. It is limited by a vision-only system
3. It does not work in all conditions
Also:
1. It took us until 2018 to make backup cameras standard; automatic emergency braking won't be standard until 2029
2. Tesla's push for FSD has created an industry-wide dragnet effect pushed ADAS, TACC, AEB, Bluecruse, EyeSight, Super Cruise, DeepSeek, etc. from competitors
3. Would you take
10% fewer accidents from driving under the influence? From lane merges? From distracted driving? From speeding? Features like seatbelts, crumble zones, and airbags are injury & death
reduction tools, not magic cure-alls. The push for FSD is the same way. Non-telescoping steering wheels were responsible for some really horrible injuries & deaths, but the technology got iteratively better over the years, despite slow industry adoption:
Though the collapsible steering column was invented in the 1930s, GM didn’t begin installing them until 1967.
The first attempt to do something about this came in 1956. Ford began installing "deep dish" steering wheels and safety belts on their car lines. The deep dish wheel collapsed about 6" in the event of a frontal crash. This wasn't enough and people were still being killed by the steering column. In the mid-sixties collapsing columns were designed and installed in cars. The government liked the statistical improvements and they've had a say so in car designs ever since.
As far as ADAS systems go:
According to the national crash database in the US, Forward Collision Prevention systems have the potential to reduce crashes by 29%. Similarly, Lane Keeping Assistance is shown to offer a reduction potential of 19%, while Blind Zone Detection could decrease crash incidents by 9%.
According to a 2021 research report from Canalys, approximately 33 percent of new vehicles sold in the United States, Europe, Japan, and China had ADAS. The firm also predicted that fifty percent of all automobiles on the road by the year 2030 would be ADAS-enabled.
Moving into the future:
1. imo LIDAR or a similar system will be required for Level 5 autonomous vehicles. Tesla stands to make
boatloads of money if they adopt a more functional system.
2. Very curious to see if you will ever really be able to auto-share your FSD personal vehicle in the future. I think Tesla sees the potential dollar signs in the future, which is why they skipped the "cheap" Model 2 variant & are moving directly into the self-driving ride-sharing market. If they can crack that code & also figure out the home robot assistant situation, they will be worth
trillions (they exceeded a $1 trillion dollar market value again this month).
2. Tesla isn't necessarily
wrong for their approach, at least not from a business perspective: customers would rather have a solution they can
afford & provides an imperfect yet
usable iterative improvement (ex. like the cruise control to TACC to Autopilot migration over the years). Consumer supervised FSD costs $8k; Waymo's sensor suite costs $100k & is only available for ridesharing. My buddy drives from CT to NYC a few times a week (4-hour round-trip) for his work commute & it can now self-drive his
entire route.
Today's iteration of consumer-vehicle Autopilot is great! However, Autopilot is
not perfect. I can't foresee the current HW4/AI4 solving L5 FSD, so either a new technology will need to be created or LIDAR will need to drop in price & get adopted for mass production. But imo
some progress is better than
no progress! In 2024, we had nearly
40,000 deaths from car accidents. For comparison, the 2003 Iraq War lasted for 7 years with about 4,500 American deaths total, or an average of 650 deaths per year, meaning we killed over 60x as many people
just driving on the road in the same time span. Which, to me, is
insane.
So yes, I will
gladly accept iterative progress for imperfect improvements, which also has the side effect of moving the
entire industry forward. Ford's BlueCruise didn't even come out until 2021 (7
years after Autopilot's first release) because
moral incentives don't drive profits.