[ad_1]
Cars
Published on November 28, 2018 |
by Michael Barnard
November 28, 2018 by Michael Barnard
Editor's note: Tesla comes tweeted that Tesla's owners have now traveled 1 billion kilometers on the autopilot.
In honor of this landmark event (no pun intended), I am once again publishing one of my favorite autopilot articles of all time, an article by Mike Barnard published in 2015. Enjoy .
Tesla recently released its autopilot mode for its cars. His intellectual approach to autonomy is fundamentally different from Google's and is superior.
One of my antecedents is robotics. I spent a year digging into PhD theses from robotics programs around the world while I was working on a startup idea for specific applications of swarm robots. We have come to software architecture, simple simulations, 3D modeling of physical robots and specific applications with tax value. I have depth here without pretending to be a robot and I continued to pay attention to the terrain from the outside.
I am therefore comfortable to say that in general, there are two approaches to robots from point A to point B.
→ The first is the paradigm of the world map, in which the robot or a connected system has a complete and detailed map of the world and a route is planned along it taking into account obstacles in advance . Basically, the robot must find its way beyond or beyond each obstacle, which generates a lot of programming.
Yes, it's a shark cat riding a Roomba.
→ The second is the paradigm of the subsumption architecture, in which a robot is first designed to survive the environments in which it will be, and then equipped with mechanisms to search for objectives. Then, without any idea of the world map, the robot goes to point B. The robot is sturdy and can stumble across obstacles without thinking about it. The original Vacuum Roomba was a pure beast of under-consumption.
Obviously, both have strengths and limitations and obviously, at least for me, a combination is the best choice, but it should evaluate the choices of Tesla compared to those of Google.
Google is starting to use the paradigm of the world map. For one of its cars to work, it takes an updated 3D model, at the centimeter scale, of the entire route taken. Google's cars are ridiculously unwieldy – by design – and when they are faced with something unusual, they stop completely. Basically, all the intelligence has to be provided by lab employees who are writing better software.
Why would Google start with this huge requirement? Well, in my opinion, without having spoken to any of the officials in the decision, it is probably because it corresponds to their biases and their blind spots. Google builds large data sets and solves the problems based on these data with intelligent algorithms. They do not build real world objects. And the split that I have highlighted above in the world map paradigms vs subsumption is a very real dividing line between academics and research around robotics. It was very easy for Google researchers and the worldview of robotics to come together and confirm their mutual biases. Others say that Google takes a low risk approach by going straight to level 4 autonomy. Although I'm sure it's part of the decision-making process, I guess it's a bit of a rationalization of their biases. The absence of Tesla's crash so far has also proved otherwise, but it's still early.
To be clear, Google cars can do what Teslas can not do, at least under the controlled prototype conditions they test. They can travel from point A to point B in the cities and regions Google has mapped to the centimeter scale, which are essentially areas south of San Francisco and some demonstration areas. You can not enter a Tesla, give it an address and sit down. This model has clear advantages in terms of the performance of the Google model compared to the current capabilities of Tesla and, although it is not trivial, it is activated by the world map model.
Tesla, on the other hand, begins with the underconsumption model. First, the car is extremely capable of surviving on the roads: high acceleration, high deceleration, high lateral rotation speed and accuracy, great survivability in the event of a collision. Then he will be more able to survive. All the car must drive on the highway is the knowledge of the lines and cars around it. Then he adds cameras to give him a clue on the proper speed. It has only a handful of survival goals: do not hit the cars in front of you, do not let other cars hit you, stay in your lane, change lanes when you are asked and your safety is assured. Because of its great maneuverability – survivability – it can have suboptimal software because it is better able to get out of bad situations. And he has a human backup.
And if this is where Tesla stopped, all those who hid their autonomy would be fundamentally correct. But Tesla does not stop there.
Tesla uses real-world intelligent search assistants to put concentrated and experienced instincts into his cars. They call the Teslas drivers. Each autopilot action and driver intervention are uploaded to the Tesla cloud, where they are combined with all the other decisions made by cars and drivers. And every driver who passes on a road automatically gets the knowledge of what cars and drivers have done before. In real time.
So, for example, a few days after the download, Teslas was already slowing down automatically for the turns they were taking at a speed previously. And do not try to take offramps marked confusingly. And do not exceed speed limits in places where the signs are obscured.
A few days after their availability, the first people across the United States gathered in less than 59 hours across the United States, with about 96% of the driving done by the car. Given the requirements of Google, they should have sent at least two cars, one or more with hyper-accurate mapping functionality, and then a day or a week later, when the data was integrated, the autonomous car itself. And there would have been no chance of hijacking or detouring for Google's car. He could literally not drive on a non-pre-mapped road at the centimeter scale. But Tesla drivers could just go.
People drive Teslas on country roads and city streets with autopilot, this is certainly not the ideal situation as a location only to which others claim that Tesla is limited. And Teslas did not hit anything; in fact, were recorded as avoiding accidents that the driver did not know. Survivability remains very high.
Tesla cars drive independently in many places where Google cars can not and will not work for years or even decades. That's because Teslas does not depend on perfect centimeter-scale cards that are up to date to be able to do anything. Subsumption wins the world's cards in a very large number of real situations.
Finally, Teslas have a map of the world. It's called Google Maps. And Tesla performs more accurate mapping with its sensors for more accurate driving maps. But Teslas does not need a centimeter precision in his world map to move. They blend well with much coarser maps that are much easier to build, store, manipulate and superimpose with the necessary intelligence. These simpler cards combined with reduced consumption will allow Teslas to move easily from one point to another. They can already go to the car park and come back on their own in controlled environments; the rest is only responsibility and regulation.
Fast jumps in autopilot capabilities just days after release should give Google a serious break. By the time his software geniuses prepare the Google car for prime time on a vast subset of roads, Teslas will be able to create circles around them.
About the author
Michael Barnard is a C-level technology and strategy consultant who works with startups, existing companies and investors to identify opportunities for significant growth in their bottom line in a transition economy producing few carbon. He is the editor of The Future is Electric, a publication of Medium. He regularly publishes low-carbon technology and policy analyzes at sites such as Newsweek, Slate, Forbes, Huffington Post, Quartz, CleanTechnica and RenewEconomy, with some of his work included in manuals. Third party articles on his analyzes and interviews have been published in dozens of news sites around the world and have reached the top spot on Reddit Science. Much of his work comes from Quora.com, where Mike is a leading writer every year since 2012. He is available for consulting assignments, speaking engagements and positions on the board of directors.
[ad_2]
Source link
