NVIDIA's automotive division has unveiled a new processor called Thor, designed for cars coming out in 2025. The company says Thor will likely be available starting with an EV from Chinese automaker Zeekr.
Tesla is the biggest name in self-driving cars, but with NVIDIA's infrastructure, it wouldn't be hard for a third party to be the first to complete LV. 4 or 5 autonomous driving capabilities. In this article, we'll show you how NVIDIA is making its mark on the self-driving car industry.
Skills needed for autonomous driving
There are a few basic skills required to build a jaunt.
- Deep learning-based recognition: Learns from vast amounts of labeled data to identify objects such as vehicles, pedestrians, traffic signs, and Detect and classify objects to understand your surroundings and make informed decisions.
- Sensor Fusion: NVIDIA DRIVE software fuses data from a variety of sensors to create a comprehensive and robust representation of the environment, improving the vehicle's perception.
- Mapping & Localization: Knowing exactly where your vehicle is located within its surroundings so it can navigate with precision and Can accurately recognize location and trajectory.
- Route planning and control: Determine safe and efficient routes of travel, taking into account factors such as traffic rules, road conditions, and the behavior of nearby vehicles and pedestrians to optimize the vehicle's behavior and ensure smooth and stable driving.
Why NVIDIA DRIVE?
Automotive AI supercomputers
In January 2015, NVIDIA released a image processing solution DRIVE PX. A year later, they released DRIVE PX2, which they claimed was the world's first "in-car AI supercomputer.
DRIVE PX2 uses NVIDIA's GPUs and deep learning algorithms to help the company's automotive partners develop autonomous solutions. NVIDIA has built automotive AI technology over the years with its powerful GPUs, represented by Thor.
Changes as development flows
The development flow has changed from model-based learning to Data-driven learning, there has been a lot of interest in the performance of GPUs recently. As a result, NVIDIA's GPU infrastructure has become a huge boon to autonomous driving development.
NVIDIA is a leader in high performance computing and image, we built an end-to-end platform for autonomous driving based on decades of experience in AI. This has enabled us to provide large-scale systems for autonomous vehicle development.
NVIDIA DRIVE provides developers with all the building blocks and algorithm stacks needed for autonomous driving. This enables them to build and deploy cutting-edge AV applications more efficiently.
NVIDIA DRIVE, The journey to autonomous driving
NVIDIA Drive Thor achieves up to 2,000 teraflops of performance and unifies all entertainment into a single architecture. Based on the 5nm Ada Lovelace architecture, it delivers significant performance improvements over NVIDIA's first-generation drive platform, resulting in greater efficiency and lower overall system costs.
Also, cloud-based simulation platform to build a We were able to generate sensor output information obtained while driving a virtual car. We built an infrastructure of data processing to enable bit-by-bit, real-time hardware-in-the-loop (HIL) development and testing.
Experts give NVIDIA's simulation platform high marks for accurately modeling the complex interdependencies of various systems in autonomous vehicle software. Within the simulation platform, everything happens in real time, ensuring timing and performance accuracy.
NVIDIA has posted strong growth in the automotive market, driven by sales of autonomous driving solutions, computing solutions for electric vehicle manufacturers, and AI cockpit solutions.
How NVIDIA DRIVE Leverages Data for Autonomous Driving
3D Mapping
When building autonomous driving simulations like the one in the video above, it's important to create a realistic representation of the driving environment. With the NVIDIA Omniverse collaboration platform, you can create content that builds highly accurate 3D environments pipeline can be created.
Not only are they similar to the driving environment, but they also have a variety of material-specific properties that are identical to how the vehicle's sensors react in the real world.
Regenerate Sensor Data
High-accuracy sensor data is required for high-fidelity development and testing after the driving experience has been realized.
The sensor model contains devices that are standard on autonomous test vehicles. For camera data, the image pipeline is based on the It works by rendering a deformed image based on the characteristics of the camera lens.
Modeling automotive behavior
Once the control signals are sent to the in-car computer, the car must ultimately react like a real vehicle. NVIDIA ensures that all movements, including details like the car's reactions, are properly mimicked in the simulation platform.
Beyond that, vehicle dynamics also play a key role in generating accurate sensor data. To learn more about how NVIDIA is utilizing data for autonomous driving, click here.
NVIDIA DRIVE SIM
NVIDIA DRIVE SIM is a tool for building autonomous driving simulations by moving real-world traffic data into a metaverse. 3D assets, scenarios, and other key components needed for simulation. These are then reorganized into simulation scenes that can be manipulated as needed.
With EDR recorders attached to more than 99% of Tesla's new cars, gateway logs stored on SD cards, and sensor data and trip logs attached to vehicles, Tesla has Collect a lot of data from users. This is how they develop their autopilot or FSD features. The reason the average company can't easily jump into the self-driving industry is because it's hard to collect Tesla-level self-driving data.
But by running the simulation through NVIDIA Omniverse, the rich graphics and AI power is available like a traditional cloud service. With Omniverse autonomous driving simulations, the expectation is that automakers that lack on-road data will be able to catch up to Tesla. With NVIDIA's provision of autonomous driving infrastructure to automakers, "catching up to Tesla" may be within reach.
Conclusion: Service features matter, not self-driving infrastructure
Tesla CEO Elon Musk predicts that his company's cars will soon be able to drive themselves without human assistance, but that prediction is actually nine years old. Experts say it's still a long way from becoming mainstream.
In addition, the entire Tesla's verticalization strategy, and many automakers are struggling with deteriorating profitability. NVIDIA DRIVE's autonomous driving platform is one such example, and while autonomous driving is not yet fully realized, automakers see NVIDIA's arrival as an enabler to focus on building autonomous data. In fact, self-driving cars powered by NVIDIA's cutting-edge technology are showing high reliability on the road. This is because the infrastructure for autonomous vehicle development is built on powerful GPUs, and autonomous driving can be simulated using real-world traffic data.
Therefore, it's time to focus less on infrastructure and more on the features and key goals of your AI service. Think about the features of the service you want to build, and consider the Focus on data design and quality.
Summary
- NVIDIA is supporting autonomous development with powerful GPUs, infrastructure for autonomous vehicle development, and simulations that mimic the real world.
- Now that the automotive industry is able to collect data for self-driving training like Tesla, we can expect to see more research and competition in autonomous vehicles.
- With NVIDIA's arrival, the self-driving car industry can focus more on the features and key objectives of the AI service, and less on data design and quality.
Reference
- Nvidia Wants to Be the Brains of Your Self-Driving Car - CNET
- NVIDIA Self-Driving Cars Are Still Training Hard Today | NVIDIA Blog
- Tesla-led self-driving technology will be a game changer : Shin Dong-ah
- Why Nvidia's AI Technology Is Essential for the Autonomous Vehicle Industry | The Motley Fool
.svg)
