Tesla’s ‘neural net’ is set to be the star of Elon Musk’s AI day (TSLA)

  • Tesla’s AI day comes just days after NHTSA launched a probe into its Autopilot system.
  • The event will provide updates on Tesla’s self-driving technology and serve as a recruitment opportunity.
  • Here’s what experts expect Elon Musk and other executives to discuss at the event on Thursday.

Tesla’s AI day, set for Thursday evening at the automaker’s Silicon Valley headquarters, could help set the record straight when it comes to Elon Musk’s grand plans for self-driving.

Tesla’s so-called neural net, which supports its Autopilot and Full Self Driving (FSD) technology, will likely be a primary focus of the event, Goldman Sachs analysts said in a note on Monday.

Musk is expected to provide updates on the company’s AI progress, but any announcements are likely to face extra scrutiny following the National Highway Traffic Safety Administration (NHTSA) announcement on Monday of an investigation into Tesla’s Autopilot system and 11 collisions that happened under its control.

In the past, Musk has said the system, which comes standard with all Teslas and enables a car to steer, accelerate, and brake automatically within its lane, makes the car safer than competitors. However, experts have said the data released by Tesla to prove the point is misleading.

On Wednesday, two US senators further upped the ante, calling for the Federal Trade Commission to investigate Tesla’s marketing practices around its Autopilot and Full Self-Driving software, dubbing it “potentially deceptive and unfair.”

Thursday’s event, scheduled for 5 pm PT, will take place nearly two years after Musk held Tesla’s first “Autonomy Day,” during which he made several lofty predictions about the future of self-driving cars. At the time, he said Tesla would have “one million robotaxis on the road” by the end of 2020. That didn’t quite pan out.

The event will primarily serve to recruit top talent, Musk said, but will also provide insight into the company’s progress toward fully autonomous cars. The event will feature a keynote speech from Musk, as well as several hardware and software demos led by Tesla engineers. Guests will also be able to test ride new vehicles, including the Model S Plaid.

Tesla’s neural net is crucial to its self-driving systems

The system emulates the functions of the human brain inasmuch as it allows the vehicle to analyze its surroundings via cameras and determine what it needs to do when it encounters obstacles by identifying and labeling different routes and images.

The system becomes more accurate the more samples it receives, especially through real-world experiences rather than simulations, Musk says. To date, the neural network is processing data from Tesla’s entire fleet — about 1.5 million vehicles.

Musk may also defend Tesla’s  AI approach that uses cameras instead of lidar sensors like other autonomous vehicles, most notably Waymo — a startup widely recognized as a front-runner in the race to self-driving cars.

The most recent beta version of Tesla’s FSD software was pushed out as an update to select Tesla drivers in July 2021 through the company’s early access program to the $10,000 or $199 per month software.

Goldman Sachs analysts also note Tesla is likely to discuss its progress towards its next edition of the software — an update that could help the FSD advance from a program that still requires a licensed operator to monitor the car’s progress to a truly autonomous vehicle that does not require constant supervision.

Project Dojo

Tesla is also expected to discuss its progress on its super-computer, project Dojo. The computer helps label and train the neural network using the video data from the cars. Dojo is said to be one of the most powerful supercomputers in the world.

The carmaker hinted in its invitation that it will provide “an inside look at what’s next for AI at Tesla beyond our vehicle fleet,” which could include advancements toward solar storage and improved efficiency in manufacturing its vehicles, the analysts said.

Go to Source