From horsepower to computing power: autonomous driving gets“boost” from SoCs

There is a saying in the auto industry that as cars become more connected, automated, shared, and electric (hence the term “CASE”), data is the new oil.

If that’s the case, then data processing or computing power is definitely the new horsepower: TOPS (trillion operations per second), rather than engine horsepower, is becoming the new performance metric.

No companies have benefited more from this change than the “Big Three” chip firms of NVIDIA, Qualcomm and Intel’s Mobileye, which all provide SoCs, and therefore TOPS, for ADAS and automated driving along with their suite of SoCs as well as full software stacks.

Nowhere was this more evident than at CES 2022 back in January, when all three made numerous announcements teaming up with major automakers like GM, Ford, Honda, Geely’s ZEEKR and Baidu’s JIDU, among others. Mobileye unveiled its next generation of EyeQ chips to take on the other two while Qualcomm CEO Cristiano Amon said his company can be considered a “transportation company.”

More recently, BMW Group, Qualcomm and Arriver announced a long-term strategic cooperation to co-develop next generation of Automated Driving Systems utilizing BMW’s current AD stack with Arriver’s Vision Perception and NCAP Drive Policy products on Qualcomm’ SoC.

There is no doubt: the “Big Three” are the winners in the era of CASE, and in particular autonomous driving, as features and capabilities expand requiring more computing power to process more data from more sensors.

To do this, you need a central computer.

One of the major announcements that came out of CES 2022 was the expansion of an ongoing partnership between NVIDIA and autonomous trucking technology company TuSimple to design and develop an advanced autonomous domain controller (ADC) specifically engineered for TuSimple’s Level 4 autonomous trucking applications.

This ADC would be that central computer unique and critical to TuSimple, as it scales its “Drive-Out” program which has successfully completed seven runs since that first milestone run last December, and eventually make it a regular occurrence on public roads come 2024.

NVIDIA was an early investor in TuSimple. The NVIDIA GPU Ventures was part of a group of investors that invested more than $20 million in TuSimple back in 2017. Over the years, TuSimple has used NVIDIA GPUs, DRIVE PX 2, Jetson TX2, CUDA, TensorRT and cuDNN to develop its autonomous driving solutions.

It looks like this collaboration has paid off for both companies as TuSimple became the first autonomous trucking company ever to put a truly driverless truck on the road.

The ADC will incorporate NVIDIA’s DRIVE Orin SoC, which is specifically designed for AI-based autonomous driving applications. The collaboration with NVIDIA is intended to accelerate TuSimple’s ability to put autonomous trucks on its proprietary Autonomous Freight Network (AFN) at scale with a production-ready computing solution capable of handling both the unique requirements and powerful computational needs of TuSimple’s autonomous driving system (ADS).

The scalable NVIDIA DRIVE Orin SoC delivers 254 TOPS of performance.

The ADC is an integral part of TuSimple’s autonomous driving system and serves as an autonomous semi-truck’s central compute unit that processes hundreds of TOPS, including mission-critical perception, planning, and actuation functions. As a result, TuSimple can expedite the development of a high-performance, automotive-grade, and scalable ADC that will be integrated into future autonomous truck production programs, such as the one NAVISTAR that has received more than 7,000 reservations.

This strategic vertical integration will give TuSimple more control over the ADC’s capabilities and accelerate its development timeline.

TuSimple will leverage its proficiency in Level 4 ADS for semi-trucks, while NVIDIA will contribute its DRIVE Orin hardware and AI expertise based on its decades of experience designing cutting-edge accelerated computing platforms. TuSimple will own usage rights to the ADC reference design, including certain limited “first-use” provisions, and intends to work with third-party manufacturers to produce the ADC.

Despite its importance to autonomous driving, the ADC is not yet commercially available in the market today. Working with NVIDIA, TuSimple is trying to change this by bringing one of the first purpose-built, high-performance, production-ready ADCs for autonomous commercial operations to market. Not only is it the world’s most advanced ADC, according to TuSimple, but it would also be engineered specifically for TuSimple’s Level 4 autonomous trucks.

The ADC streamlines the electronic and software architecture by reducing the complexity of computations. It receives information from sensors such as cameras, LiDARs and radars as well as other sources, processes the data and constructs a digital representation of the surrounding environment. The ADC software then makes policy and planning decisions surrounding the actions the vehicle should take, and eventually controls the critical functions based on the planning decisions.

It basically is incorporated within TuSimple’s ADS by taking in all the data and input from the environment collected by all the sensors and combine these data with high definition mapping to construct a digital representation of the truck’s surroundings, makes policy and planning decisions to determine appropriate driving decisions, and then sends signals to the truck’s engine, steering, braking and other functions, which are actuated accordingly.

This vertical integration of the ADC into the ADS is expected to help TuSimple mitigate the risk of relying solely on a nascent supply chain. It will also enable fast iterations and development of different concepts and generations. The ability to iterate faster, because of its direct relationship with NVIDIA, relative to standard industry practice, should allow TuSimple to reach an optimal design earlier and, ultimately, deploy the solution within its AFN faster.

Today, speed is of the essence and computing power is king. The ADC takes care of both for TuSimple as it accelerates the scaling of autonomous freight runs.