Tesla is doubling down on its focus on becoming the global leader for large-scale AI applications such as self-driving cars and humanoid robots:
“While many are pulling back on their investments, we are investing in future growth
โ including our AI infrastructure, production capacity, our Supercharger and service
networks and new products infrastructure โ with $2.8B of capital expenditures in
Q1. […]
The future is not only electric, but also autonomous. We believe scaled autonomy is
only possible with data from millions of vehicles and an immense AI training cluster.
We have, and continue to expand, both. To make FSD (Supervised) 5 more
accessible, we reduced the price of subscription to $99/month and the purchase
price to $8,000 in the US.”
Tesla has been making some seriously cool strides in both AI software and hardware, aiming to bring us closer to the dream of full vehicle autonomy and even a ride-hailing service. They’re betting big on a “vision-only architecture” which means they’re using only camerasโno other sensorsโpaired with some pretty advanced neural networks that learn from a vast trove of real-world driving data.
Quick Overview of Tesla’s AI Efforts
Here’s a condensed bullet-point overview of Tesla’s AI efforts:
- AI & Robotics: Developing scalable autonomy for vehicles and robots using advanced AI and efficient inference hardware.
- Tesla Bot: Creating autonomous bi-pedal robots to perform repetitive or unsafe tasks, focusing on software for balance, navigation, and interaction.
- FSD Chip: Designing AI inference chips to maximize performance-per-watt for Full Self-Driving software.
- Dojo Chip: Building AI training chips to enhance performance and efficiency of the Dojo system.
- Dojo System: Constructing the Dojo system to improve machine learning compute power, featuring a public API for broader access.
- Neural Networks: Advancing deep learning to process real-time data from Tesla’s fleet for tasks like perception and control.
- Autonomy Algorithms: Crafting algorithms for accurate world representation and trajectory planning using combined sensor data.
- Code Foundations: Optimizing Autopilot software integration with custom hardware for enhanced performance and reliability.
- Evaluation Infrastructure: Developing tools for performance tracking and realistic testing using anonymized fleet data.
Full Self Driving (FSD)
Earlier this year, they launched FSD (Full Self-Driving) V12 in a supervised setup, and itโs been a game-changer, reinforcing Tesla’s approach towards scalable self-driving tech. They’re not stopping there, though. Tesla plans to beef up its core AI capabilities even more in the coming months.
You can see FSD 12.3.5 in action here — 40 minutes of almost flawless driving in the city! ๐
In terms of hardware, they’ve completed the upgrade to Hardware 4.0. This new in-car computer boasts better processing power for AI and enhanced cameras to feed all that crucial visual data into Tesla’s neural nets.
On the software side of things, Tesla is pushing the envelope with a 30-day free trial of FSD for eligible U.S. and Canadian cars. This version of FSD can handle tasks like lane changes, navigating forks in the road, and maneuvering around obstacles.
Plus, they’ve rolled out an Autopark feature for existing FSD and Enhanced Autopilot customers.
Ride Hailing and Robotaxi
Looking ahead, Tesla is diving into ride-hailing, promising to integrate this service right into the Tesla app. They believe that their software experience is unmatched and that the integration will be seamless.
I have to say, the app interface really looks fresh and clean and I can already see myself ordering Robotaxis (or integrating my Tesla car into the robotaxi network to earn revenue):
By the way, check out my Robotaxi simulation game for fun and learning:
Tesla Bot
Interestingly, Tesla didn’t mention any update on the Optimus humanoid robot in their most recent earnings report. Well, Elon did say in the earnings call that the Tesla bot will be able to perform factory work by the end of this year (2024).
๐ Tesla Bot Optimus: Is $5,000 per Share (TSLA) Realistic?