Tesla is betting the farm on computer vision which works through its eight cameras and ultrasonic sensors – nothing else.
Tesla has announced that it has stopped equipping its cars with a forward fig radar which is a critical component of its AutoPilot system. This change in hardware is currently reflective for only the Tesla Model 3 and Model Y, but in time one can expect a change to happen to the Model X and Model S as well. For the longest time, Elon Musk has been a proponent of just relying on cameras augmented with advanced AI and computer vision technology.
Musk has in fact gone out of his way to disparage LiDARs which are in other cars that tout self-driving technology, made famous especially by Waymo, so they are the visual equivalent to radar, allowing for spatial awareness for the car. However, in the recent past, Tesla has made some big investments towards getting rid of any kind of radar system.
It has developed its own self-driving chipset jettisoning Nvidia from its cars. At one time Tesla had also hired Jim Keller, the former AMD, Intel and Apple executive who had helped build their CPUs. All this while, Tesla has also been testing its “full self-driving” software in a beta test with employees and select customers. This software has been trained on a supercomputer called the “Dojo”.
As of now, Tesla isn’t fully done using the radar. It is just limiting and disabling features for a limited amount of time. For instance, Autosteer is an AutoPilot feature that can keep the car centred in the lane and even around curves – but now this will be restricted to 120 km/h. Tesla is also restricting the longer minimum distance between cars for the following purposes.
Its smart summon feature which allows the user to hail the car from a parking lot may get disabled at delivery. Even emergency lane departure avoidance will get disabled. Tesla says it will restore all these features in a couple of weeks.
Tesla is not removing the radar from its lower volume vehicles Tesla Model S and X. It says it is focusing on the Model 3 and Model Y because it sells far more of them which will help it to tune the system faster because more sales will enable it to capture more self-driving data.
“Transitioning them to Tesla Vision first allows us to analyse a large volume of real-world data in a short amount of time, which ultimately speeds up the roll-out of features based on Tesla Vision,” the company said in a blog post.
As of now, Tesla hasn’t announced anything about the radar sensors in its cars in China or the Model S and X. Recently, it has introduced new models of Model S and Model X, but their production has been delayed.
Radar sensors are actually quite common in cars – they help detect fast-moving objects even in poor visibility and also help power ADAS features like automatic emergency braking. Modern cars have an assortment of sensors including cameras. Automakers rely on levels of redundancy but Tesla is betting the farm on AI and computer vision.
Musk has been critical of LiDARs because of their cost. He has been talking up the vision-based system his team has been developing with eight cameras and 12 ultrasonic sensors which are passed through a neural network in real-time.
In the recent past, Tesla’s AutoPilot system has also been rated at best level 2 autonomous. It has also been criticised for its marketing of the feature which lulls the consumer into a false sense of security which has caused many accidents, including numerous fatalities. Tesla’s confidence in its self-driving beta and criticism of AutoPilot could be accelerating the shift to what it is calling Tesla Vision
Our AI-based software architecture has been increasingly reliant on cameras, to the point where radar is becoming unnecessary earlier than expected,” Tesla wrote.