AI and Its Effects on Autonomous Vehicle Storage

By Martin Booth

The volume and velocity of data on our roadways globally have never been higher, and the ability to leverage and transform this massive amount of data into real-time intelligence and value is critical.

  • TAGS

While today's cars typically implement about 16 to 64 gigabytes of internal data storage, on average, mostly to accommodate map and infotainment functions, tomorrow's autonomous vehicles are expected to require even more capacity. Internal research leads us to believe that autonomous vehicles may soon need more than one terabyte of storage to support a broader list of advanced capabilities that include intelligent driver assistants, voice and gesture recognition, eye-tracking, driver monitoring, black box recording, cognitive capabilities (that learn and analyze driver preferences to improve on them), and connectivity with other vehicles and traffic infrastructures. The volume and velocity of data on our roadways globally have never been higher, and the ability to leverage and transform this massive amount of data into real-time intelligence and value is critical.

More computing power is coming to the car to process data from a myriad of sensors, algorithms and connections to the outside world, resulting in more and more data being captured and analyzed. The data needs to be processed, and some of it will need to be stored locally or uploaded to cloud storage, as discussed in the following three areas: autonomous driving, digital assistants and data center connections.

While these areas promise to have dramatic impact on automotive storage themselves, they become particularly compelling when you factor in one of the most exciting trends in automotive today: artificial intelligence (AI). A computer-based technology that enables system applications to perform human-like tasks, AI not only promises to literally drive autonomous vehicles of the future, but has already been leveraged in many areas by auto manufacturers to automate and improve driving experiences. From assisted driving and digital voice assistants to deep-learning capabilities and connections with data centers and infotainment systems, AI-supported vehicle applications collect massive amounts of data, requiring new strategies to store and manage that information, both locally and in the cloud.

Autonomous Driving
Autonomous vehicles can collect 750 megabytes of data per second from their surrounding environment through a variety of sensors, such as cameras, radars and LiDARs (light detections and ranges), that help them to steer, brake and accelerate through traffic. In-vehicle sensors read, compare and physically map data to its environment so the vehicle can recognize and respond to obstacles that may come in its path. This generates massive amounts of data to successfully maneuver the vehicle—all of which requires a lot of additional storage capacity.

The compressed and processed data is then compared to a high-definition (HD) map in order to derive an accurate vehicle position. Such maps reside on top of the standard map data and contain information such as lane markings, curbs and signs that can easily double the map size. This information is used to produce real-time, actionable insights that determine how the car will navigate through traffic.

Some data, such as the vehicle's "drive" data, may need to be saved for a number of days or months, depending on regulatory, operator or original equipment manufacturer (OEM) requirements. The drive data recordings can range from seconds, for black-box accident recording, to days, relating to the monitoring of fleet vehicles for insurance, predictive maintenance, and other purposes, resulting in a wide disparity of storage requirements. Even if the drive data is uploaded to cloud-based storage, a local copy needs to be available. All of this contribute to greater on-board storage capacity.

It is expected that most autonomous cars of the future will be equipped with the latest in Wi-Fi connectivity and vehicular communications enabling passengers to browse the Internet, send and receive emails, and even watch a downloaded movie. Data requiring long-term storage will eventually be uploaded to the cloud. However, for moving vehicles, strong connections within an automobile will not always be guaranteed—plus, data must be stored locally, which adds to the storage density requirements.

Digital Assistants
Extending autonomous driving to new levels, digital assistants perform AI functions through software algorithms that generate an incredible amount of data. Unlike intelligent personal assistants that provide voice services for mobile devices, digital assistants enable the vehicle to learn about drivers' personal preferences, interests, driving style and more. This enables the system to not only provide a driving experience based on personalized information, but also continually increase its knowledge, imitate human behavior by analyzing behavioral patterns, and interpret real-life driving scenarios just like a human—even taking over the wheel when required.

This form of machine learning (ML) is actually a subset of AI and was developed through science, so when applied to machines or devices, it will think and act almost like humans do. In AI, machines or devices execute tasks that humans consider smart. In ML, machines or devices are given data that they learn from and, through deep learning (DL) practices, enable many of the AI activities. Deep learning helps to break down tasks into manageable chunks and mimics the activities so the system can learn on its own. The more it learns, the more data it generates, and the more storage capacity that is required.

Data Center Connections
The connected car continually moves closer to enabling smartphone-like experiences and includes several categories of systems that cover infotainment, road and traffic warnings, vehicle diagnostics, navigation and more. AI-based infotainment systems enable drivers and passengers to receive and send emails, perform Internet searches and interact with smartphone applications—all through voice commands—and transmit many megabytes of data per second, depending on what application the vehicle is running at the time. As more automotive applications become available, and online-enabled, data will need to be quickly and efficiently moved between local vehicle storage (as soon as it is collected) and the cloud. This rise in automotive data generated will require considerable local storage and cloud gateway buffer coding.

Vehicle safety is one of the most important contributions that AI is making to connected cars. Through vehicle-to-vehicle technology driven by wireless connectivity, connected cars will have the ability to communicate with one another by informing other vehicles around them of what they are doing. For example, if a driver fails to slow down while approaching a red light, the connected car could alert cross traffic to avoid an accident. Additionally, connected cars will have the ability to interact with roadway infrastructures, such as traffic lights and signs, through vehicle-to-infrastructure technology. A simple example is a traffic light telling a connected car it is about to turn red so the car will know to slow down.

Flash-based Storage
Storage is a critical piece of the overall automotive solution and represents a significant part of the bill-of-materials. As the car increasingly becomes a data center on wheels, with multiple computers onboard and interconnecting through the cloud, storage optimization will become critical to ensure performance and reliability. In some of the recent infotainment systems, software updates can now be cached and infotainment apps can be buffered to reduce network bandwidth use at peak.

To address the need for higher performance, higher capacity, lower latency, and better reliability and endurance, many automobile makers are turning to flash-based storage for storing the operating system and advanced software applications, for collecting and analyzing drive data recordings, for buffering cloud communications (also for bandwidth optimization), and for storing local copies of infotainment data.

Now proven in the rigorous automotive environment, flash storage supports the high-capacity requirements of autonomous cars, and is available in highly compact packages that are smaller than a U.S. penny. In AI-enabled and autonomous vehicles in which the complexity of systems increase, yet the real estate is limited and every inch of the car counts, the flash-based storage form factors fit into small systems and consume minimal physical space within the vehicles themselves.

Local, on-board vehicle storage that supports connected cars and autonomous driving must be able to withstand harsh environments and perform reliably in the vehicle for long lifecycles. Given the varied environments in which automobiles operate (hot and cold extremes, wet and dry conditions, smooth and bumpy surfaces, shock and vibration challenges), local in-vehicle storage will soon have quality and reliability requirements that far exceed the storage requirements in smartphones or other mobile devices, and may soon approach the requirements found in mission-critical enterprise storage.

Final Thoughts
The reliable collection and storage of data are foundational in how automotive designers can incorporate AI and achieve safe and reliable autonomous drive in cars. Though we won't see driverless vehicles on our highways and byways this year, we will continue to see test cars, and a steady and dramatic progression among auto makers, technology companies and network infrastructure providers to put the pieces together prior to mass deployment. And though fully autonomous cars are still a ways off, and AI continues to infiltrate our daily lives in other ways, advancements in storage will go hand in hand with its success. It will become more apparent that data, and how data is used and stored in the connected car of the future, will be key in realizing the exciting AI visions of these manufacturers.

Martin Booth is the director of automotive solutions marketing at Western Digital Corp.