Evaluating the Use of Enterprise HPC Hardware in Autonomous Trucks

October 18, 2022

By Braden Cooper, Product Marketing Manager

In today’s high-performance computing technological landscape, there is a stark contrast between the compute capabilities of enterprise hardware in datacenters and low-performing embedded systems in edge and transportable environments. The autonomous trucking industry in particular has seen a growing need for higher performance compute, storage, and networking solutions in its rugged edge environments. While size, weight, and power (SWaP) optimized embedded systems are the long-term objective of the industry, current autonomous trucking development platforms require access to high levels of compute to avoid any bottlenecks imposed upon their critical software developments. When looking to solve the developmental compute need of the autonomous truck industry through enterprise hardware – there are some important factors to consider.

1. Meeting the Present and Future Compute Demand
   
   The first consideration is the dataflow map of the systems within the autonomous truck. When on the road, the truck will be capturing data from several cameras, a LiDAR system, mechanical sensors, and other data sources. For real-time decision making, the data must be rapidly captured, processed through the inferencing model, and the resulting decisions sent to corresponding control systems throughout the vehicle. In current autonomous truck development, there is also typically a data storage element – whether capturing the entirety of raw data over the course of a trip or storing critical data when an anomaly or incident is detected by the inferencing model. In order to keep all of these aspects of the dataflow operating efficiently – there must a be a balanced input and output to each phase of the system. For example, if the cameras, LiDAR, and other data inputs are gathering several gigabytes of data per second, the computing and storage elements of the system architecture must be able to handle that capacity of data flow without introducing a bottleneck.
   
   Existing embedded compute and storage solutions can solve the problem, but only to an extent. As the resolution and quantity of sensors increases, embedded platforms are quickly outpaced, leading to a tradeoff in the incoming data integrity and ultimately the success of the system architecture. Outside of the edge embedded space, enterprise class compute hardware is more than capable of scaling alongside the data needs of autonomous trucks. Although integrating enterprise hardware in edge environments has its own challenges, the key is to identify the throughput needs of current development platforms and the throughput needs of production systems into the future, avoiding compromise on the data input or effectiveness of the AI model.

    

2. Size, Weight, Power, and Ruggedization
   
   Secondly, a primary gap between commercial off-the-shelf electronics and those that are integrated in autonomous trucks is their SWaP profile and capacity to operate in extended environmental conditions. In a datacenter, a typical 2U to 4U server with a depth of 20” or more is commonly used for GPU or FPGA scaleout. These systems are designed to operate in air-conditioned environments that experience little to no vibration or shock loads over their life. Systems integrated into autonomous trucks conversely often require much smaller mechanical profiles and a rugged design to meet the rigors of the outside world. When considering the use of enterprise electronics in an autonomous trucking environment, the mechanical domain must be carefully defined. While typical commercial servers may exceed the allowable size or weight, transportation optimized rugged servers can provide a good middle ground solution. Systems designed from the ground up for the environmental parameters of autonomous trucks can provide maximum compute performance in a mechanically optimized package.

    

3. Industry Standards, Safety, and Qualifications
   

   
   Lastly, as the autonomous vehicle market in general continues to develop, the governing bodies and regulatory agencies continue to define laws regulations, and qualification requirements for the vehicles. The impact on leading autonomous truck integrators will be massive in the coming years as regulations change and mature with vehicle experience and new findings. When it comes to adopting high-performance computing hardware into autonomous trucks, certain elements must be carefully considered such as the flexibility and modularity of the compute system designs. As industry standards are defined, modifications to the electronics and the corresponding regulatory testing of the electronics is a key expertise to have. Over the next several years, the autonomous truck compute integrators which can navigate shifting compliance standards most efficiently will have a strong advantage over their competition. This efficiency will come from flexible system designs, robust qualification test plans and execution, and a strong understanding of current and future regulatory standards.

Ultimately, integrating enterprise high-performance computing electronics into autonomous trucks is an important step in the current development cycle of the industry. By using datacenter level compute capacity, software developments can accelerate without constraints or bottlenecks in the system architecture. Integrating mechanically optimized rugged servers also mitigates the challenges introduced by edge environmental conditions of autonomous trucks. This hardware also provides a long-term roadmap in terms of scalable compute performance as new technologies are introduced over the next several years. OSS offers a product portfolio of compute, storage, and networking systems optimized for the autonomous trucking industry’s current and future needs.

Click the buttons below to share this blog post!

Return to the main Blog page

Leave a comment

Name *

Email *

Comment *

---

Also in One Stop Systems Blog