Tesla Shares Details of its New Self-Driving Chipset at ‘Autonomy Investor Day'
【Summary】Tesla held its “Autonomy Investor Day” today at the automaker's Silicon Valley headquarters to provide analysts and investors with updates about the automaker’s autonomous driving hardware.
Tesla held an "Autonomy Investor Day" today at its Silicon valley headquarters to provide analysts and investors with updates about the automaker's autonomous driving hardware. Musk first tweeted about the event on April 13, writing "On April 22, Investor Autonomy Day, Tesla will free investors from the tyranny of having to drive their own car."
Much of today's presentation was focused around the technical specs of Tesla custom-built self-driving hardware or system-on-a-chip (SOC). Tesla's head of Autopilot Pete Bannon provided an overview of the new chip that will power Tesla's camera-based Autopilot autonomous driving system.
The new chip is designed exclusively for Tesla's autonomous driving functions. The SOC contains six billion transistors and can process 2.5 billion pixels per second. It's got a batch size of one, meaning that as each image is processed by the GPU as soon as it's generated.
Tesla claims the new SOC is the "world's most advanced computer for autonomous driving" and offers a substantial performance increase over Tesla's current hardware. It can process over 144 tera operations per second (TOPS).
Tesla CEO Elon Musk boasted that Tesla's new chip is even better than the NVIDIA DRIVE Xavier SOC, which can theoretically process 21 TOPS.
"As a car company that never designed a chip, how could Tesla manage to design the best in the world?" Musk said, boosting of the automaker's accomplishment. Musk added that the automaker's SOC will offer "a new level of safety of autonomy in Tesla vehicles."
"A year from now we'll have over a million cars with full self-driving, software, everything," he said. He said it was improbable but true that Tesla had designed the best self-driving chip. One reason, he explained, was that Tesla wanted a chip only for autonomous driving, while others, like Nvidia Corp, wanted a chip that could do other things as well.
The chip is also designed to use less power. Bannon said that it was important for the chip to use less than 100 watts of power. He said the chip consumes 72 watts of power, with 15 watts of that powering Tesla's neural networks.
The new chip is being manufactured by Samsung and is small enough to fit behind the vehicle's glove box. "It does not take up half of your trunk." Musk said.
For an extra layer of safety, the SOC includes two independent operating systems, each with their own with DRAM memory, flash storage chips, and power supplies.
That makes them highly redundant, Musk said. "Any part of this could fail, and the car will keep driving."
The chips is already being installed in Tesla models during assembly. Musk said that "All Teslas being produced right have now everything necessary for full self-driving, we just need the software to improve."
Musk said that all Model 3's produced over the past ten days already have the new self-driving computer installed, and Model X and Model S vehicles had it since March.
Tesla is Not Relying on Lidar for its Autonomous Driving System
Although most developers of autonomous vehicles use lidar for navigation, Musk does not think lidar is necessary.
"Lidar is lame" Musk said today, referring to the laser technology that most competitors are using to develop self-driving cars. Musk explained that humans drive with vision only, so Tesla designed their autonomous driving hardware to do the same—using advanced computer vision and deep neural networks.
The chip will process data from eight vision cameras, 12 ultrasonic sensors and radar on the vehicle combined with a neural network. Bannon said the neural network's primary function is visual recognition.
Tesla CEO Elon Musk said that all Tesla vehicles being produced now have the chip and Tesla is about halfway through the design process for its next generation chip, which Musk estimated would be about three times better than the current system.
Tesla has been working on a self-driving chip since 2016 and previously used hardware from Nvidia.
Unlike Nvidia's DRIVE hardware, Musk said that Tesla's self-driving computer fits behind the glovebox, "it does not take up half of your trunk."
Musk said that Tesla will offer retrofits to current Tesla owners who purchased the optional $5,000 "full self-driving" package in the next few months. Later this year, the system will be able to recognize traffic lights and stop signs, and perform automatic driving on city streets, according to Tesla's website.
When asked by an analyst just how much money Tesla is spending to develop its Autopilot autonomous driving system, Musk said "It's basically our entire expense structure." without provided an exact figure.
Today's event comes just two days before the electric automaker reports its Q1 earnings. Tesla is expected to announce a quarterly loss on declining deliveries of the Model 3 sedan, the automaker's first mass market electric car.
Originally hailing from New Jersey, Eric is a automotive & technology reporter covering the high-tech industry here in Silicon Valley. He has over 15 years of automotive experience and a bachelors degree in computer science. These skills, combined with technical writing and news reporting, allows him to fully understand and identify new and innovative technologies in the auto industry and beyond. He has worked at Uber on self-driving cars and as a technical writer, helping people to understand and work with technology.
Ford is Testing a New Robotic Charging Station to Assist Drivers of EVs With Disabilities
Ford Raises the Prices of the F-150 Lightning Electric Pickup Due to Rising Raw Material Costs
The BMW 7-Series to Feature HD Live Maps From HERE Technologies for Hands-Free Highway Driving in North America at Speeds up to 80 MPH
AutoX to Use the 'Eyeonic Vision Sensor' from California-based SiLC Technologies for its Robotaxi Fleet in China
LG Develops ‘Invisible’ Speaker Sound Technology That Could Revolutionize In-Vehicle Audio
Researchers at South Korea’s Chung-Ang University Develop a ‘Meta-Reinforcement’ Machine Learning Algorithm for Traffic Lights to Improve Vehicle Throughput
Zeekr’s New 009 Electric Passenger Van is the World’s First EV to Feature CATL’s Advanced ‘Qilin’ Battery With a Range of 510 Miles
Redwood Materials is Building an Electric Vehicle Battery Recycling Facility in South Carolina
- The BMW 7-Series to Feature HD Live Maps From HERE Technologies for Hands-Free Highway Driving in North America at Speeds up to 80 MPH
- NHTSA Upgrades Tesla's Autopilot Investigation to 'Engineering Analysis'
- Tesla Believes Its Dojo AI System Will Help It Win the Self-Driving Car Race
- Volvo-backed EV Battery Maker Northvolt, Founded by Former Tesla Execs, Raises Another $1.1 Billion to Help Fund its Expansion
- Stellantis to Idle its Illinois Assembly Plant Indefinitely, Citing the High Costs of Electric Vehicles
- SiLC Technologies Launches its Eyeonic Vision System, a LiDAR Sensor That Can Identify Objects up to 1 Kilometer Away
- Researchers at South Korea’s Chung-Ang University Develop a ‘Meta-Reinforcement’ Machine Learning Algorithm for Traffic Lights to Improve Vehicle Throughput
- Toyota Offering Multiple Solutions for Non-Drivable bZ4X Electric SUVs
- Baidu CEO Believes That SAE Level-4 Autonomous Driving Systems Will the First to Enter Commercial Use After L2, Skipping Over L3
- Tesla's Battery Supplier CATL Unveils its New ‘Qilin’ Battery That Can Deliver 600+ Miles of Range to EVs