Originally Published on Linkedin 3rd October 2022

 

 Photo by Tesla
Photo by Tesla

Tesla AI day 2022 was a major event for the company, demonstrating its advances in artificial intelligence over the past year. What we can learn from Tesla AI day 2022?

The event showed that Tesla is making significant progress in artificial intelligence, both in terms of software and hardware.

This included:Demonstration of their Optimus Robot (last year it was a man in a suit)

  • Progress on FSD (Full Self-Driving) Software
  • Dojo Supercomputer

The key takeaway from the event is that Tesla is creating a scalable way to train it’s fleet of vehicles which will have a wider impact than just cars. Their powerful AI models can be repurposed for different uses which is highlighting the power of AI and how quickly new solutions can be created. We are starting to see this happening in other sectors using general-purpose AI models. This is where I believe we will see the biggest impact of AI in the coming years, in terms of transforming industries and creating new ones.

Optimus Robot

Optimus Robot has come a long way in the past year. Some may argue that this isn’t hard given that last year when the robot was first unveiled as an idea, it was a man in a suit. Tesla cars are essentially robots on wheels, so it is not surprising that the company would be able to transfer this technology to a robot on legs. What is impressive, however, is the speed at which they have been able to progress.

Whats changed with Tesla Bot

In just one year, the Optimus Robot has gone from being a man in a suit to a fully functioning robot with its own independent fingers and hands. This is thanks to Tesla’s use of their autopilot software and hardware, which they have been able to adapt for the robot. Not only does this show the potential of Tesla’s artificial intelligence, but also the company’s ability to quickly adapt and innovate. They still do have a long way to go before the robot can be commercialised, but they are making significant progress.

No alt text provided for this image

Their Optimus Robot is now based on the Tesla autopilot, using sensors to identify its environment. It can now do some rudimentary tasks like carrying objects.

The second version of this Robot will have hands which have independent fingers that are opposable and can grip like a human hand. This is thanks to the addition of metal tendons.

They are confident they can mass produce these at a high volume eventually at a cost of around $20,000.

Elon Musk discussed what could that mean for the economy. He believes that it has the potential to transform the economy.

He said that in the near future, we will see a practical use case for the Optimus Robot in the Tesla factory. The goal is to have the robot work alongside humans in order to increase efficiency and productivity. This is an ambitious goal, but one that Tesla seems confident they can achieve in the coming months or years.

 

Progress on FSD (Full Self-Driving) Software

In addition to the hardware advances, Tesla has also made significant progress on its FSD software. This is the software that has been touted as eventually enabling Tesla cars to drive autonomously, however, this has been proving to be more difficult than initially thought. Musk has previously stated that they had hoped to have this software ready by now, but that it has been much harder than they anticipated. He did say, however, that he is confident they will get there eventually and AI day 2022 showed a path to how this could be achieved through the ability to train AI models and scaling the amount of data that they can process from both a growing fleet of vehicles and from simulation.

No alt text provided for this image

Last year they had 2000 customers using their FSD software in Beta. This year they have increased that to 160,000 customers. They are now processing a vast amount of data per month. To get to the point where their cars can drive themselves, they need to be able to process even more data and continue to learn from it.

The most impressive aspects of what was discussed were:

Showing how FSD has improved in specific interactions and can make “human-like” decisions. For example, when a Tesla makes a left turn into an intersection, it can choose a trajectory that doesn’t make close calls with obstacles like people crossing the street.

No alt text provided for this image

Tesla trains the company’s AI software on its supercomputer Dojo, which is one of the world’s most powerful computers for training AI models. They are improving the software so that it can handle more complex situations like construction zones and bad weather.

No alt text provided for this image

Once training is performed this information is fed into customers’ vehicles over the air. They are able to process video feeds from Tesla’s ever growing fleet of multi camera equipped vehicles on the road today. Tesla have a simulator built in Unreal Engine that is used to improve Autopilot. They showed how quickly these models can be generated based on data they can call on from their fleet of over one million vehicles.

 

Dojo Supercomputer

No alt text provided for this image

One of the most impressive things that Tesla showed at AI day was their Dojo Supercomputer. This is a Tesla designed supercomputer that is designed specifically for training artificial intelligence models. It is an impressive piece of hardware, and it is clear that Tesla is serious about becoming a leader in artificial intelligence. The Dojo Supercomputer is able to train models much faster than other computers, and it can also handle much larger datasets. This is important because the more data that an AI model can learn from, the better it will perform.

Tesla has greatly reduced the training times using Dojo which will help them continue to scale. This is thanks to the Dojo Supercomputer’s ability to parallelize the training process.

In addition to being able to train models faster, the Dojo Supercomputer is also more efficient.It uses less power and takes up less space than other supercomputers. The downside of all this power is it is so powerful that it tripped the power grid in Palo Alto.

Tesla may even be able to offer Dojo as a service much like Amazon Web Services or Microsoft Azure which could potentially be a pretty big offering in itself.

Conclusion

Tesla has made great strides in artificial intelligence over the past year. With advances in both hardware and software. Tesla is creating a scalable way to train its fleet of cars which will have a wider impact than just cars. Their powerful AI models can be repurposed for different uses which is highlighting the power of AI and how quickly new solutions can be created. We are starting to see this happening in other sectors using general-purpose AI models. This is where I believe we will see the biggest impact of AI in the coming years, in terms of transforming industries and creating new ones. 

Yash Varma