The Imminent Winner of Autonomous Driving
Autonomous driving will be the next major innovation in transportation.
We can liken this breakthrough to the change from horse-back riding and horse-drawn carriages to the use of automobiles in the early 20th century. Autonomous driving will bring convenience, freedom, and safety.
And not everyone will even need their own autonomous car.
A fleet of autonomous robotaxis will bring the convenience of ride-sharing services (Uber/Lyft) at a far cheaper price point, with greater safety, and without the drama of human conflict or fraud.
What is required for
To answer this question, we do not need to dive straight into the complex engineering and advanced technology needed for autonomous driving (LiDAR, radar, cameras, digital maps). We begin at the simplest starting point: the example of a human driver.
Though human driving is the current standard, it has significant limitations.
Vision can be blurred or blocked. Can only focus on direction.
Brain can be preoccupied or tired. Decisions may not always be consistent
Execution might be hindered. There will always be a lag in reaction.
And this is where Tesla changes the game.
We believe that Tesla is approaching autonomous driving from first-principles perspective.
Let’s take a deep dive into their solution.
Tesla's use of cameras has further benefits. See why we believe this technology will give Tesla an edge in the race to autonomous driving.
Only Tesla has this data advantage because it is drawing from its existing fleet of one million+ Tesla cars. Tesla is gaining this large amount of varied, real world data from its fleet.
What about Lidar
LiDAR is a great sensor. However, several papers, one from 2018 before Tesla’s Autonomy Day presentation and one from 2020, outline that current deep learning technology can recreate the depth data that LiDAR provides, rendering LiDAR largely obsolete. LiDAR is not on the critical path for autonomy.
Therefore, Tesla's decision not to use LiDAR is a business decision that will ultimately help Tesla win the race to autonomy.
Paradoxically, having fewer sensors allowed Tesla to create vehicles at scale while being able to offer greater safety and accuracy.
redundancy for safety
The principle here is that any part of this could fail and the car will keep driving. You could have cameras fail, you could have power circuits fail, you could have one of the Tesla FSD computer chips fail, [but the] car keeps driving.
The probability of this computer failing is substantially lower than someone losing consciousness.
Tesla Autonomy Day, 1:17:52–1:18:17
Data Learning Engine
Deep neural network training AI to make smarter and smarter full self-driving decisions
Tesla uses its large amount of varied, real-world data to train and improve its autonomous driving software through an iterative learning process
How it all works
Inaccuracies and driver interventions are detected (the example here is driving in lighted tunnels at night)
Tests are run on this isolated incident to figure out the problem
More examples of similar situations are gathered from the fleet
The neural network is trained based on all this data
The new training is deployed to the fleet
This process is repeated as many times as necessary to get it right
Tesla's ability to get real-world, varied data while developing their custom full self-driving computer gives them an edge above their competitors.