Tesla has been releasing FSD Beta updates on a semi-regular schedule, about once every two weeks. The latest version is 10.6.1, but it was only rolled out to select beta testers. Most FSD testers are still on beta 10.5.
Tesla's FSD Beta on a Model S
Black Tesla/YouTube
According to Elon Musk, Tesla is set to release FSD Beta 10.8 this week, skipping over version 10.7.
However, it looks like v10.7 was released internally, as the technical release notes have now been leaked to Sawyer Merritt on Twitter.
FSD Beta 10.7 is version 2021.44.5.5. The technical release notes state:
- Improved object attributes network to reduce false cut-in slowdowns by 50% and lane assignment error by 19%.
- Improved photon-to-control vehicle response latency by 20% on average.
- Expanded use of regenerative braking in Autopilot down to 0mph for smoother stops and improved energy efficiency.
- Improved VRU (pedestrians, bicyclists, motorcycles, animals) lateral velocity error by 4.9% by adding more auto-labeled and simulated training examples to the dataset.
- Reduced false slowdowns for crossing objects by improved velocity
estimates for objects at the end of visibility.
Reduced false slowdowns by adding geometric checks to cross-validate lane assignment of objects.
- Improved speed profile for unprotected left turns when visibility is low.
- Added more natural behavior to bias over bike lanes during right turns.
- Improved comfort when yielding to jaywalkers by better modelling of
stopping region with soft and hard deadlines.
- Improved smoothness for merge control with better modelling of merge
point and ghost objects positioned at the edge of visibility
- Improved overall comfort by enforcing stricter lateral jerk bounds in
trajectory optimizer.
- Improved short deadline lane changes through richer trajectory modeling.
- Improved integration between lead vehicle overtake and lane change gap selection.
- Updated trajectory line visualization.
There are a lot of changes in this FSD release. Based on the release notes it sounds like this release may help fix some of the larger complaints testers have had, mainly around false slowdowns and jerkiness when taking a turn.
It looks like there will also be an updated visualization for the car's trajectory, which we will be sure to add to our full list of FSD Beta visualizations.
Tesla is set to release beta 10.8 with the holiday update as early as this Tuesday. Between the FSD Beta release and the new holiday features, it's likely to be the most anticipated and exciting release of 2021.
While we haven’t even seen AI5, Tesla is already thinking about its next set of chips that will power Tesla’s future FSD computers. In a set of posts on X, Elon Musk outlined the future of Tesla’s AI hardware, stating that its “strategic importance is hard to overstate.” He also revealed that Samsung, the current manufacturer of Tesla’s AI4 chip, will produce the next-generation AI6 chip at a new, dedicated fabrication facility in Texas.
While AI4 is already used in Tesla’s Robotaxis, which are practically driving autonomously, it’s hard to imagine where the company will be with a computer that’s many more times as powerful. Things like the recent video of a Tesla stopping at a toll booth, waiting for the driver to pay, and then taking off will become commonplace, not only at booths but also in various nuanced scenarios, such as someone waving you by, or stopping for a friendly neighbor that waves you down.
Samsung’s giant new Texas fab will be dedicated to making Tesla’s next-generation AI6 chip. The strategic importance of this is hard to overstate.
Samsung currently makes AI4.
TSMC will make AI5, which just finished design, initially in Taiwan and then Arizona.
To understand the need and brilliance behind why Tesla is doing what it's doing with AI6, we need to take a look at the problem it is being engineered to eliminate. Currently, Tesla, like many others, faces a real two-language problem in its AI development.
The vast majority of its AI training is done on a supercomputer cluster, Cortex, which is built around Nvidia’s powerful H200 GPUs. However, the vehicles that run the software use Tesla’s custom-designed hardware - HW3/AI3 or HW4/AI4. This means that every time the AI model is improved - whether the vision neural nets or the ones that determine distance, or any of the many little parts that come together to build FSD as a whole - there’s a whole second step. It must be developed and validated on Nvidia’s architecture, then rebuilt and re-validated to run on Tesla’s own AI hardware.
Tesla runs AI4 chips side-by-side with the H200 GPUs in Cortex to minimize this, but that still means there is a step in between the training getting completed and the actual model being able to run on a vehicle. This slows down the feedback loop from training to deployment, even if Tesla has built a Universal Translator to help move FSD from one piece of hardware to another.
AI6 Solution: One Chip to Rule Them All
AI6 is the definitive solution to this problem. While the upcoming AI5 chip (to be made by TSMC) represents a performance leap over AI4, AI6 will be a transformative leap in architecture.
The key innovation of AI6 is the direct integration of Tesla’s Dojo Supercomputer chip architecture into the same hardware that will be in Tesla’s vehicles and Optimus. The goal is to closely integrate the training and vehicle hardware.
By having the same Dojo architecture in the data center for training and in the vehicle for inference, Tesla will have a single, unified hardware pipeline. The two-language problem vanishes, drastically simplifying Tesla’s development process and enabling a much faster pace of innovation and updates.
Manufacturing Collaboration
The plan for AI6 goes beyond just chip design. Elon’s announcement revealed that Tesla plans to build a strong manufacturing partnership with Samsung. After all, much of Tesla’s expertise is focused on the machines that build the machines.
Samsung has agreed to allow Tesla to assist in maximizing manufacturing efficiency at the new Texas fab. It’s unusual to see this kind of relationship between the fabricator and the client. Elon doubled down on this - he wants to walk the fab lines to personally accelerate the pace of progress.
Samsung agreed to allow Tesla to assist in maximizing manufacturing efficiency.
This is a critical point, as I will walk the line personally to accelerate the pace of progress. And the fab is conveniently located not far from my house 😃
Tesla isn’t just designing the chip, but helping to make sure it’s made in the most efficient and optimal way possible, all while leveraging Tesla’s experience and Samsung’s fab skills in one facility.
Fully Integrated
Tesla’s AI6 announcements come as a surprise since Tesla is still a year out from releasing AI5, but it shows that Tesla plans to integrate even more of its AI stack. From influencing the manufacturing process of the chip itself to designing a unified hardware architecture for both training and inference, Tesla is building a self-reinforcing ecosystem.
This approach will create a feedback loop that allows for rapid improvement. While AI5 will be a big step up from the current generation, AI6 is revolutionary as it becomes Tesla’s major AI platform.
At the recent X Takeover event this past weekend, two of Tesla’s most important leaders gave in-depth interviews that provided a unique view into Tesla’s path forward and how everything comes together internally.
An interview with Elon Musk laid out the grand vision for Tesla and his other companies, focusing on the ambitious “what and why.” Later, Head of Vehicle Engineering Lars Moravy provided the more grounded, engineering-focused “how,” detailing the immense work it takes to turn the vision into a reality.
What emerged from these conversations was a clearer picture of Tesla’s strategy: a relentless, long-term vision for the future of transportation, AI, robotics, and energy, supported by a world-class engineering team capable of developing the processes to turn these products into a reality.
Robotaxi & Cybercab
For years, the concept of Unsupervised FSD, as well as Robotaxis, has been the focal point of Tesla’s future. In his interview, Elon provided fresh details on the way they expect the business model to work. Tesla plans to operate a fleet where some vehicles are company-owned, while others are owned by customers. This is essentially a combination of Uber and Airbnb, taking a bit of a hybrid approach between the two different styles.
He also went on to confirm that the purpose-built, two-seater Cybercab would complement, but not replace, Tesla’s existing models. This is key, because many have thought that Tesla would end their consumer vehicle sales or drastically reduce them as they transitioned to an AI services company, and became less of a car company. Now, it seems we know that they’ll have a lasting stake in personal car ownership.
The Cybercab, which is a revolutionary vehicle without driver controls, requires an equally revolutionary process to build it. In his interview, Lars Moravy provided the answer and detailed the unboxed manufacturing process that Tesla has been developing.
The unboxed method challenges a century of established automotive assembly by breaking the vehicle down into smaller, parallel sub-assemblies, allowing more work to be done simultaneously. The goal is to drastically shorten the main assembly line, enabling vehicle sections to be built in parallel and come together at the end.
Lars also noted that Tesla has already done the initial batch of crash testing for the Cybercab prototypes, and the vehicle has passed with flying colors. This isn’t surprising for Tesla, which integrates vehicle safety right into the structure of the vehicle, building castings that transfer force away from occupants.
The Semi
While Elon’s interview focused on some of his grander ambitions like Mars colonization, Lars provided some tangible updates on two of Tesla’s most anticipated vehicles.
On the Tesla Semi, Lars confirmed that progress is well underway at the Semi factory in Reno, Nevada. After years of focusing on engineering prototypes to ensure the reliability of a commercial workhorse vehicle, Tesla is now expected to ramp up production by the end of 2025, continuing through into early 2026. The business case for the Semi is crystal clear - build a no-brainer choice for shipping and logistics companies, who need to weigh the initial buy-in and infrastructure costs against operating costs per mile.
The Semi, just like other EVs, absolutely trumps diesel trucks in cost per mile, due to lower energy costs and less maintenance. However, the somewhat hidden advantage here is that truck drivers drastically prefer to drive the Tesla Semi over other diesel trucks, citing things such as better visibility, a smoother ride, and easier driving. These are advantages that could lead to improved employee retention and easier driver recruitment.
Meanwhile, the Semi simply needs to have infrastructure installed at the starting and ending locations for major delivery companies, enabling end-to-end supply chain handover.
The Roadster
Lars also talked about Tesla’s upcoming Roadster, confirming that it’s still in development, with the team preparing for a mind-blowing demo sometime soon. Elon previously hinted at this demo during a visit to the Tesla Design Studio, where he said a mind-blowing demo would be coming by the end of the year.
The goal for Tesla is to make it the last, best driver’s car before the world begins transitioning to full autonomy. Lars also touched on one of the biggest challenges with the Roadster. There is an immense engineering challenge being taken on now - and it's the SpaceX package. This package is set to use cold-gas thrusters to push the Roadster past what is conventionally possible. In fact, just as Elon has previously mentioned, the Roaster may be able to “fly a little.”
Last, best driver’s car
Lars Moravy
You can watch the full interview below. Lar’s portion on the Roadster starts at 26:30.
Optimus: Sustainable Abundance
One of the most ambitious parts of Elon’s vision is the Optimus humanoid robot. He has stated his belief many times that the robotics business could be many times more valuable than Tesla’s entire automotive business, and if it works as planned, it definitely will be.
The current Optimus V3 design is intended for volume production, with Elon foreseeing a future market of billions of humanoid robots - not made just by Tesla, but the market as a whole. That many units could simply eliminate human poverty and usher in an age of sustainable abundance.
That grand vision is built on top of the manufacturing and automation expertise that Lars’ team is pioneering every day. With volume production of Optimus to begin next year, and real work already being done in Tesla’s factories, we may see humanoid robots making a real impact on the lifestyle and livelihood of people within the next few years.
The Unfair Advantage: Getting Sh*t Done
All of these ambitious ideas and products are enabled what what is perhaps Tesla’s true sauce - its unique internal culture of getting sh*t done. Lars’ interview provided us with a rare look inside to see just how it all comes together.
He described working with Elon as unique - the discussions are grounded in physics, and Elon trusts his teams to turn his dreams and ambitions into reality. This, in turn, creates a culture of mutual respect and high expectations.
The collaborative spirit extends to the relationship between engineering and design, which Lars described as highly unusual for the auto industry. Rather than the two teams being hostile to each other, they work together to make bold design and engineering choices, like the Cybertruck, into reality.
Underpinning all of this is what Lars himself calls Tesla’s superpower: in-house automation and manufacturing engineering teams. These teams work to design the machine that builds the machines - innovating and solving problems at a level and speed that is simply not possible when relying on external vendors.
This combination of a relentless long-term vision, alongside a first-principles engineering culture, allows Tesla to take big risks and make big plays that define its future path. While all of Tesla’s timelines are ambitious, these interviews make it clear that the ambitious vision is paired with a concrete and innovative plan for execution.
![No Driver Needed: Tesla FSD Stops at Toll, Waits for Driver to Pay and Takes Off Again [VIDEO]](https://www.notateslaapp.com/img/containers/article_images/fsd-beta/fsd-toll-booth.webp/57e008aff60c6c1e998e304eca200cda/fsd-toll-booth.jpg)
