Advanced driver assistance systems (ADAS) and the promise of autonomous vehicles have led to visions of use cases, from driverless long-haul trucking to robo-taxis. McKinsey predicts the ADAS market could reach $35 billion in revenue by 2021;[i] however, making these use cases a reality is requiring significant advances in enabling technologies, including the internet of things (IoT), artificial intelligence (AI) and machine learning (ML).
Scaling ADAS Platforms with Strategic Infrastructure and Architecture Decisions
In this TechTarget white paper find out how a strategic infrastructure and IT architecture support the ADAS platforms.Read More
Autonomous vehicles also have critical digital infrastructure implications, which industry participants are encountering and solving as they go. An early challenge, for example, was data ingress, or how to collect the data generated by each test vehicle. Autonomous driving systems must also be able to aggregate, analyze and distribute that data, as well as data from other sources such as traffic and weather information, in real-time – with all the necessary security and privacy controls in place. And as the degree of autonomy advances (from level 1 for some driver assistance to level 5 for fully autonomous), the amount of data to aggregate and analyze will continue to soar. Current test drives for L2 autonomy are generating up to 20 terabytes (TB) of data a day, while more advanced sensor sets for higher levels of autonomy (L4 and above) may generate up to 100 TB/day. Moreover, engineers and trusted third parties across the globe may need to access this data which can cause challenges in terms of data access and distribution. This means manufacturers must find a way to minimize data transfer latency by establishing proximity between datasets and accessing sufficient compute resources to manage the data on a global scale.
The need to establish this type of infrastructure means centralizing data on-premises is no longer an option. Distributed data processing requires digital infrastructure (compute + storage) that is connected by an efficient global communication fabric. To accomplish this, companies are finding they must deploy hybrid infrastructure at well-connected locations that can deliver high-speed, secure access to edge devices, multiple clouds, private data centers and on-premises data, data brokers and partners. This is driving the development of connected vehicle ecosystems based on third-party partnerships and hyperconverged infrastructures as shown in the diagram below.
Understanding the evolving data requirements for ADAS
Advanced driver assistance systems require a range of data-intensive capabilities, including the ability to manage data volume, variety and veracity while minimizing latency. The systems must be able to overcome the data challenge of four key control points:
- Sensors: Sensors collect data and use infrastructure connectivity to support both vehicle-to-vehicle and vehicle-to-infrastructure systems.
- HD Mapping: HD mapping includes high-precision 3D and other necessary information to enable autonomous vehicles to see the changing world around them and detect safety hazards.
- Processors: Processors, such as engine control units (ECUs) and microprocessor units (MPUs), process the data and make essential decisions related to navigation, fuel economy, preventive maintenance and more.
- Software: Software systems, especially those using AI and ML, provide the functionality for each new autonomous driving use case.
Image source: McKinsey[iv]
From an evolving infrastructure point of view, another key challenge for autonomous vehicles is Hardware-in-the-Loop (HIL) testing, which vehicle manufacturers use to verify and validate the software in the ECUs. HIL is a significant technology challenge because of the huge datasets that must be stored, managed, transmitted and analyzed. Around the clock operational support is required to optimize data ingest to the cloud or to on-premises facilities. In addition, the test rigs require frequent physical access, which means they must be located in facilities that make this frequent access practical, easy and cost effective. And, as companies invest in more complex HIL architectures running in potentially different facilities and locations worldwide, ensuring that this distributed data is synchronized is critical.
The only way to support the technology demands of autonomous driving systems is with low-latency access to digital ecosystems of clouds, networks, hardware/software providers and partners for high-speed, secure exchange of data such as those found on Platform Equinix®.
Hardware-In-The-Loop Testing on Platform Equinix
Continental’s connected vehicle ecosystem delivers global performance and scalability
When Continental Automotive wanted to extend its leadership in autonomous driving solutions, it needed to evolve its digital infrastructure to support faster, globally scalable AI processing. To accomplish this, Continental established a successful ecosystem of partners that enabled the company to build what is essentially its own global supercomputer – that is, a parallel file system capable of meeting the high-speed demands of AI and protecting sensitive data.
Continental’s ADAS Vision employs a sensor-equipped test fleet that drives over 9300 miles (15,000 kilometers) per day, generating and recording over 100 terabytes of data, which is then ingested, processed, selected, assessed and annotated, and used for training and validation of the system. The system relies on NVIDIA DGX servers for deep learning and training artificial neural networks to detect what is happening in any given scenario and decide how the car will respond. To speed development and time to market, Continental required high performance access to this data and a powerful storage solution capable of reading hundreds of thousands of images per second into the GPUs.
Continental partnered with systems integrator SVA, IBM and Equinix to build a new storage infrastructure in an AI-enabled NVIDIA DGX-ready Equinix International Business Exchange™ (IBX®) data center in Frankfurt, Germany where there is a growing ecosystem of industry peers. This new solution includes a multimode GPU cluster, non-blocking InfiniBand network infrastructure, IBM ESS with fast Non-Volatile Memory express (NVMe) drives, NVIDIA DGX servers, and NVIDIA V100 Tensor Core GPUs. Continental is also using IBM Spectrum Scale with its Kubernetes container environment for modern application development.
“As a result of our new infrastructure, we can now run 20, 40, 80 GPUs simultaneously to really speed up our training,” says Balazs Lorand, PhD, Head of AI Competence Centre, ADAS@Budapest, at Continental. With this new infrastructure, Continental achieves 14 times more deep learning experiments per month and has reduced training time from weeks to days
Thanks to the flexibility and global scalability of Platform Equinix, the solution will support growth in any direction – in containerized hybrid cloud environments, on premises and in multiple data centers. Equinix IBX data centers, located in 63 markets in 26 countries on 5 continents, offer the highest levels of operational reliability, with an industry-leading uptime track record of greater than 99.9999%, as well as access to dense partner ecosystems, including more than 1,800 networks and 2,900+ cloud and IT service providers. This means Continental can build exchange points in proximity to the data and services required for ADAS and will be able to easily meet its infrastructure needs for years to come as its autonomous vehicle use cases mature and expand.
To learn how to build a scalable ADAS Platform, download the TechTarget white paper: Strategic Infrastructure and Architecture for Scalable ADAS Platforms
You also may want to watch/read:
- [Webinar] CVST trends and strategies
- [Webinar] Emerging CVST architectures
- [White paper] Artificial Intelligence: From the Public Cloud to the Device Edge
- [Industry resources] Digital-Ready Solutions for Automotive