3. November 2018 of Zachary Shahan
There have been a couple of major Tesla autonomy announcements in the past week or so. First and foremost, Tesla pushed out a Navigate on Autopilot function. Basically, the update makes the Tesla cars go from onramp to off-ramp without having to enter the driver – just based on the route you have in the Tesla navigation system, and what magical Tesla engineers have packed behind the touchscreen.
Recently, Elon Musk told us that you can soon "walk your Tesla" as if it is a dog and control it as a child's remote-controlled car. These fun and useful improvements will come via an update to the Summon feature.
It's cool to explore the possibilities for each of these improvements and get excited, but it's somewhat more dramatic to wait in the shadows just between these two announcements. There is something that snags hidden between the parking lot and the highway. Basically, we are a few autonomous roads away from the Tesla vehicles that are fully-powered cars.
Think about it: The car can navigate a parking lot alone to get to you. It can also drive from onramp to off-ramp on the highway without having the intervention (well, once you have confirmed that it can do what it thinks it should do at critical times during the trip). What's in the middle of the two moods? In some situations – say a hotel just off the highway and a home or workplace nearby – then there's almost nothing left to block door-to-door autonomous driving. In more complicated situations, the car basically needs to find through cities by following common traffic rules and whatever freak events life can throw it on.
This is also where Tesla's approach to self-propelled technology separates the battery-solving software from companies commonly considered competitors in this space, such as Waymo (aka Google / Alphabet). Tesla uses a fundamentally different system. Waymo's self-propelled system is based on precise mapping and just knowing all that is about traffic rules and common traffic events in a relatively small geographic area. The Tesla self-drive system is based on the cars that learn to drive like humans and learn to answer road rules and surprises in the same way a person would. In other words, it's about running more like a human being than a robot with very good maps.
This gets a bit confusing and after talking with self-propelled tech experts, I've discovered that even expert outsiders don & # 39; I do not have a good picture of what Tesla does behind the scenes. We know there is shadow training in progress – Tesla's autonomous driving systems learn from their human drivers every day – but how much do they learn and what details are they aware of? We know that Tesla uses neural networks to speed up and improve the learning process and that new hardware will make these neural networks much better, but we do not have much clarity in the practical stages of these neural networks, the challenges they are TESLA Autopilot has traveled 1 , 48 billion miles (updated estimate): https://t.co/ZOfZn6HlFY pic.twitter.com / 19Ze0t7Bft4
– Lex Fridman (@lexfridman) October 9, 2018
] Updated data on total Tesla vehicle delivery of Autopilot hardware: 443k total Teslas, 279k version 2, 114k version 1, 49k without AP hardware. @elonmusk https://t.co/v34JLgLIyg pic.twitter.com/J0xukW3nMt
– Lex Fridman (@lexfridman) October 8, 2018
Recent data from Tesla and MIT indicate that the Tesla Autopilot has raised 1.5 billion miles. It was also recently revealed that Waymo's self-propelled vehicle has driven 10 million miles. It's been a while since I took a math class, but I remember billion is much bigger than million .
As I said above, these systems are not really the same. They use fundamentally different methods to pursue a self-driving future. But the thing about Tesla's system is that it teaches people how to drive like humans – except safer thanks to better attention tension, faster reaction times and better vision. Apparently, the system is now good enough to handle parking and highways. As the autonomous miles continue to grow exponentially, expects Tesla to unlock the few remaining points between the parking lot and the highway.
At the last quarterly conference call, Elon Musk demonstrated that Tesla's automotive automation software dwarfs the hardware in some "comparable" cars. It's ready to make the "little" leap from driving around the car park and driving on the highway to driving anywhere . Perhaps people who paid for Full Self Driving at the time they bought the cars became a little impatient (the same happened when Autopilot was rolled out first), but some annoyance about the waiting time is likely to be wiped out as soon as Tesla sends the Over-the-air update as allows owners to download and use the first versions of Full Self Driving . When the same drivers are retrieved from work a few years later or have a car to make money for them by driving people around while they are bored, I think they have long forgotten that they had to wait a few months or even a year to eat them First fruits of their Full Self Driving purchase.
You can remember it back in July 2016 Musk showed that the Tesla Autopilot system needed at least 1 billion miles of vehicle driving to get out of beta mode. However, it was not clear exactly how many miles over 999 million Tesla would need. "With less than 1 mile, it's simply not enough data. 1B is a necessary, but not necessarily enough, condition," Musk tweeted da.
Are we there yet? My understanding is that the answer is basically yes, or almost.
Apart from the purely technological aspect and when Tesla feels clear, it is of course a rule that must activate different levels of self-drive including full self-driving without a human tutor (aka driver) who is awake and able to take over if necessary . The technology may be ready next year, but when will you actually be allowed to send your car to pick up some avocados or airline passengers around town for some extra money?
We will see and the chances are high. Elon will break the news via a tweet.