5 Key Takeaways from Tesla AI Day - Vision, Neural Net, Dojo, Tesla Bot, and Recruitment:
Tesla AI Day is the third such presentation day that Tesla has done. They previously did Autonomy Day (2019) and Battery Day (2020). The goal of such events is to provide insights into different technologies Tesla is developing and to convince more people to join their efforts.
Tesla is head-on addressing some of the toughest challenges faced by humanity today, as the transition to sustainable energy, autonomy, scale manufacturing of batteries, and safe development of AI.
These events are a great way to really understand how Tesla is progressing as a company. The technology innovation at Tesla is very fast and the pace of it is only increasing. It is difficult for others to catch up with Tesla. They already have a couple of years of lead and it is increasing rapidly.
Most people look at Tesla as an electric car company. But Tesla has evolved over the years and they not only make electric cars today but they also make solar panels, battery storage systems for retail and commercial use, batteries that go into storage systems and their cars, computer chip hardware, battery management software, and self-driving software.
While creating these solutions, and being vertically integrated, they are effectively able to make use of learnings from different processes that go into building great products, at scale and at low cost, with optimized use of resources.
Tesla already uses robots and automation in their factory up to a great extent, while improving the finished product output and simultaneously improving the factory as well. As Elon says they are optimizing the machine that makes the machines. Tesla has also started manufacturing their own batteries, given that the demand for batteries is going to increase multiple magnitudes of order in the future.
Taking their learning from the development of self-driving AI for their vehicles over the years, it is natural for them to implement those technologies for solving different problems.
One such application of AI seems to be a human-like robot, that is able to do dangerous, repetitive, and boring tasks. Elon Musk announced that Tesla Bot is their next product and they will potentially have a prototype by sometime next year.
That said, the main goal of Tesla AI Day was to encourage people interested in solving complex real-world AI problems at the hardware or software level to consider joining Tesla. They did it by showing how they have developed their self-driving software and hardware, their approach to different problems, the challenges faced along the way and how they overcame them, and what they want to do in the future.
Let us look at the 5 key takeaways from Tesla AI Day:
Computer Vision:
Tesla has been developing self-driving technology with the help of radar and cameras for a couple of years. And eventually, they moved to a vision-only approach, given the reasoning that is how humans drive, using vision.
This approach uses neural networks to enable the car to function anywhere on earth. It moves around, senses its environment takes action intelligently and autonomously based on what it sees. The system is called full self-driving (beta) and is being tested in actual customer cars and improving rapidly.
It takes data from 8 cameras placed around the car and then it goes through the neural networks and it creates a 3D representation of its surrounding. This enables the car to accurately make decisions and subsequent actions.
Neural Networks:
For the Tesla FSD system to work efficiently and to allow the car to take actions based on what it sees, the system has to be trained for all different possibilities that can happen in the real world.
Since there are more than 1 million Tesla vehicles on the road today, they provide real-world data which helps train the neural network. But there are certain scenarios that do not happen often but the system still needs to be trained for it. For example, at times of blur vision, the immediate response at the time of the crash, pedestrians, unexpected behavior of other drivers, etc.
For training the system for such unique cases, the neural net is provided with simulated data that can be used to improve decision-making under such instances. It also allows for improving accuracy by predicting all types of obstacles on the road which are even beyond the vision of the camera or human eye.
Now, in order to rapidly and continuously improve the system, its neural network needs to be trained with a massive amount of data both real-world and from the simulator. For that, there has to be a dedicated computer to do that efficiently. Hence Tesla created the Dojo computer.
Dojo — AI Training Computer:
Tesla has previously developed their own chip for a full-self-driving computer that goes into the vehicles. Building on those learning and with the goal of being vertically integrated, they further developed their own chip called D1 Chip for Dojo computer.
This is an application-specific integrated circuit (ASIC) that is dedicated to neural network training and is capable of handling vast amounts of camera image data and is multiple times faster than other computing systems.
Multiple D1 Chips can be combined together with power & control and heat systems to create a Training Tile. And these multiple training tiles can be combined together to form the fasted AI computer for neural net training.
This can be used to train different neural nets apart from self-driving AI.
Tesla Bot:
At this point, Tesla has already talked about computer vision, neural net, simulation, new chips, and Dojo computer. Now they started talking about how these advanced technologies can be used in applications beyond vehicles.
The most logical one seems to be Tesla Bot. It will use the full-self-driving computer and cameras for vision to perform tasks that a human being is least likely to do.
Tesla is arguably one of the leading robotics companies in the world. They already have semi-autonomous robots on the road in the form of cars. They have made significant progress in full-self-driving hardware, software, inference engine and will continue to evolve. They understand the nature of the real world through computers and how to navigate through them.
Thus it makes sense to use all these technologies in a humanoid form. They also have experience with sensors, batteries, and actuators which are essential for such a product.
The prototype version of Tesla Bot is expected to be ready by sometime next year. It is intended to be friendly, be able to navigate through a world built for humans, and can eliminating danger, and perform repetitive and boring tasks.
Even though the bot can have much more computer power than a human, it will have a low max speed of 5 MPH.
The bot will have a screen on its head area that will show some useful information, it will have human-level hands. It will have multiple cameras for vision and full-self-driving computers. And it will make use of the same tools as that of a self-driving car.
Recruitment:
Throughout the presentation, the entire Tesla team mentioned that there is a massive amount of work and complex problems that need to be solved. And their fundamental bottleneck in moving ahead fast is limited human-resource. They need more people that have the knowledge and are interesting in solving real-world AI problems to join them.
Conclusion:
Tesla showcased the way they have developed full-self-driving technology using cameras and in-house-built computer chips and neural network training systems that enabled them to create one of the most advanced vision AI.
And now they are going to continuously improve for accuracy and potentially implement it in other applications, Tesla Bot being one of them. They need more smart people to join their efforts in solving these problems.
I highly recommend that you check out the entire presentation on Tesla’s YouTube channel. It will be worth it.
Thank you for your time. I hope you enjoyed reading this article. Do let me know your thoughts and any questions in the comments below.
Please like, share, and subscribe for more SpaceX, Tesla, and Bitcoin content.
You can also follow my creator account @ The Space Road on YouTube/Instagram/Facebook where I will be posting more clips and some key information about the event.
