Industry progress

Anyone who drives regularly knows that most journeys are uneventful. The real challenge is not the 99% of trips that go as planned, but the one unexpected moment that does not.

That remains one of the biggest challenges facing automated driving (AD) systems. Recently, the UK government published the findings of its consultation on new safety principles for automated vehicles. Among the proposals is the expectation that authorised self-driving vehicles should demonstrate a level of safety equivalent to, or higher than, that of a careful and competent human driver. That is a demanding benchmark. Autonomous systems are no longer being judged solely on how they perform during routine journeys but also in other situations that fall outside normal driving patterns.

That is why autonomous driving remains such a difficult problem to solve because the real world rarely sticks to the script. The industry has made enormous progress; vehicles can already perform tasks that would have seemed unfathomable just a decade ago, such as airport transit or London’s first driverless bus. Yet every unusual scenario creates another situation that must be understood, tested and validated before autonomous systems can be trusted to operate at scale.

Why simulation and every mile matters

This is where the conversation changes. A fleet manager might judge a vehicle by the miles it completes. Autonomous vehicle developers look at those same miles and see something else entirely: data. Every autonomous vehicle records a detailed account of what it sees and experiences on the road. Every junction, lane change, near miss, unexpected manoeuvre and weather condition adds to a growing pool of information that engineers use to improve and validate the technology.

Today, to improve the testing before a vehicle hits the road, AI is being implemented to use video data to train and test the AD systems. This greatly reduces the amount of road miles needed to deem it safe. These AI-driven tests can simulate corner cases that would take a lot of driving to encounter in real-world driving. Nonetheless, testing on roads in real-world environments has its place. A vehicle encountering temporary roadworks for the first time might generate valuable information. Individually, these events may seem insignificant. Collectively, they help build confidence that autonomous systems can cope with the countless situations they will eventually encounter.

Developers take a two-pronged approach. AI to train their autonomous systems and then capturing anomalies on the road that is then used to additionally train the models and update them remotely. That creates a significant burden for developers. A system that performs well during testing must still prove it can cope with many future journeys, many of which will contain situations it has never encountered before. Every deployment, software update and unusual event adds to the volume of information that must be reviewed, analysed and validated before autonomous systems can be trusted to operate more widely.

The overlooked infrastructure challenge

When people discuss infrastructure for autonomous vehicles, the conversation often focuses on what sits outside the vehicle. Charging networks, connectivity and roadside systems all have an important role to play.

Inside the vehicle, however, different infrastructure challenges are emerging. Modern autonomous systems rely on AI models that continuously interpret information from cameras, radar, lidar, maps and other onboard sensors. Those decisions need to be made in real time, which means the data they depend on must be available instantly. Relying on cloud connectivity alone isn’t enough. Latency, coverage and resilience all demand that much of this information is stored and accessed locally within the vehicle.

This is where advances in automotive storage become increasingly important. Autonomous driving systems continuously draw information from cameras, radar, lidar, high-definition maps and AI models to understand the world around them. Those decisions need to happen in real time.

Automotive-grade flash storage solutions are designed to support this type of challenge and are crucial to making sure data is available and reliable when it’s needed providing real-time in-vehicle storage for the AD computer to overcome latency and connectivity issues that may arise when accessing the cloud. Rather than simply providing more storage capacity, they help ensure AI systems can access the data they need quickly and reliably, allowing advanced driver assistance and autonomous driving functions to make decisions without unnecessary delay. As vehicles become more software-defined, storage is becoming an active part of system performance rather than simply somewhere to keep data.

Just as importantly, the workloads themselves are changing. Modern vehicles are generating, recording and updating far more information than previous generations, making it more essential for OEMs to understand how storage is being used in real-world conditions.

What fleet operators should expect next

Fleet operators are unlikely to wake up one morning and find fully autonomous vehicles ready to replace every van, truck or company car in their fleet.

The more likely starting point is in environments where operators can control as many variables as possible. For an industry built around data, every mile matters. Every journey completed, every unusual event recorded and every software update deployed adds to the body of evidence needed to support broader adoption.

Though some may continue to focus on the latest demonstrations and technological breakthroughs, behind the scenes, much of the industry’s effort is focused on a less visible challenge: collecting, storing and validating enough real-world data to prove autonomous systems can cope with the unpredictability of everyday driving, meaning automotive flash storage is likely to help determine how quickly autonomous vehicles become a practical reality on the roads we use every day.