Tesla AI Day: From Cars and Chips to Robots

Tesla is no longer a car company but a technology company. It is looking to build highly disruptive, efficient and intelligent products leveraging the whole gamut of horizontal technologies, from alternative energy to artificial intelligence. The focus is on having scalable solutions with great user experience by effecting a full vertical as well as horizontal integration of technologies and businesses.

Tesla AI Day: Tesla Strategy Counterpoint

Tesla CEO Elon Musk has been showcasing this vision through events such as ‘Tesla Battery Day’, ‘Tesla Autonomy Day’ and ‘ Tesla AI Day’. Musk wants to tell the stakeholders how the company is working on cutting-edge technologies to build futuristic life-changing products with “clean tech” as a foundation.

At the recently concluded ” Tesla AI Day”, the company unveiled the Dojo D1 chip and a humanoid robot, which are expected to open new business opportunities for Tesla. The company also showcased how it is providing training to neural networks to make better decisions in real-life driving situations.

Full self-driving (FSD) and beyond

Tesla AI Day: Tesla FSD Counterpoint

Tesla FSD can complete a trip without any action by the driver. This year, Tesla launched an FSD beta version 9.2 with an additional one-time payment of $10,000 and a $199 monthly subscription package. This powerful supercomputing capability will not be limited to Tesla cars. Any interested carmaker or tech company can join hands with Tesla as it will require more investment to build system infrastructure. When asked whether Tesla would keep autopilot as open source, Musk said it was open to licensing driverless technology to other carmakers.

Tesla cars with autopilot experienced only 0.2 accidents per million miles driven in 2020 while the US average is 2 accidents per million miles driven, which means Tesla cars with autopilot are nine times safer than other cars.

Even if we compare fire incidents, one Tesla car caught fire for

every 205 million miles (~330 million km) travelled while the US average was one car catching fire for every 19 million miles (~31 million km) travelled during 2012-2020. Public misconceptions and inappropriate comparisons make it difficult for the company to uphold the actual facts.

Analyst take: Tesla is fully committed to active and passive safety of a car, which differentiates it from other brands. But since fresh cases are coming where its autopilot failed to prevent crashes, Tesla will have to work a lot on FSD and autopilot systems before becoming a fully autonomous vehicle company. On the way, it will also have to clear investigations by government bodies.

In the long term, Tesla Robotaxis will try to reduce power consumption by improving powertrain efficiency, thus minimizing the cost for customers and reducing carbon footprint.

Tesla’s vision system and solving problems through simulation

Tesla AI Day: Tesla Vision Counterpoint

Tesla’s vision system is equipped with 8 cameras, 12 ultrasonic sensors and a powerful computer to provide safety during the journey. Tesla AI head Andrej Karpathy explained that a vision-based system works on neural networks which can function in any environment via the autopilot system. Currently, Tesla’s vision system can detect objects, traffic lights and lanes.

Now, Tesla is performing over one million evaluation runs per week to improve neural network training. It is also constantly working on increasing the GPU cluster size. Currently, Tesla has more than 10,000 GPUs, constituting the fifth-largest supercomputer in the world.

Analyst take: Tesla is keen to solve computer vision problems in real-life environments through series of evaluation testing. We hope it will ultimately help Tesla to accurately predict the behavior of pedestrians and vehicles, and prevent accidents.

Tesla isn’t only building a smart car, it also takes care of the vehicle’s cybersecurity. The company organizes the Bug Bounty program and Pwn2Own research competition to take feedback from those who are passionate about digital security. This continuous improvement in product development is one of the major pillars of Tesla’s success.

Unveiling Dojo chips

Tesla AI Day: Tesla Dojo Counterpoint

Tesla unveiled the Dojo AI training chips built on a 7nm process node and packaged with the BGA packaging technology. Tesla claims it has GPU-level compute, CPU-level flexibility and twice the network chip-level IO bandwidth which can outperform NVIDIA systems.

Tesla uses a distributed computer architecture to address latency and bandwidth as a primary optimization process. This chip reaches 362TFlops of BF16 or CFP8 and 22.6 TFlops of FP32 with 400W TDP. The power density is similar to the NVIDIA A100 HGX GPU.

Analyst take: Tesla is trying to bring 10x improvement in performance for the next generation of chips. Dojo is a ground-breaking technology, but we aren’t sure whether it can beat others in terms of performance until it comes under production. The recent semiconductor shortage will be another challenge for Tesla in solving supply chain issues for hardware development.

Tesla tends to go in for in-house design and development. We have seen how Tesla has built its battery ecosystem. In another case, it was sourcing SoCs from NVIDIA, but it started making its own self-driving SoC in 2019. The in-house AI training chip will help Tesla improve the Dojo supercomputer’s performance with reduced footprint and cost.

Tesla bot: Tesla to enter humanoid robot market

Tesla AI Day: Tesla Bot Counterpoint

Tesla is working on building a humanoid robot named Tesla Bot, which is expected to come as a prototype next year. Tesla Bot is 5’8’’ in height, 128 lbs in weight (~56.7kg), can walk 5mph (~8kph), deadlift 150 lbs (~68kg) and has a head that displays useful information. This robot is capable of performing tasks that are repetitive, boring or unsafe for human beings. Tesla is using in the robot technologies that are similar to the ones used in its cars, such as an autopilot camera, FSD computer, neural network and Dojo supercomputer.

Analyst take: We aren’t sure whether Tesla will be able to build a prototype by next year as it has a record of missing deadlines. However, Tesla Bot will eliminate jobs in the future and people will have to skill up for newer technologies or look for alternative income sources. At the same time, we believe that physical work will be a choice in the future.

Tesla’s efforts to look beyond vehicles will find a result in the introduction of Tesla Bot. Whenever it comes into production, it will bring disruption to the manufacturing, mining, retail and hospitality industries.

Tesla is yet to comment on the recent probe ordered against its autopilot system, which could have provided some confidence to the investors, regulators and customers. We think Tesla needs to educate stakeholders at first to remove misbeliefs about disruptive technologies like the Dojo supercomputer, FSD and humanoid robot.

Related Posts