Autonomous mobility: it’s a jigsaw puzzle

Building autonomous mobility is akin to sitting down to a 1000-piece jigsaw puzzle. The rubber won't meet the road until the manifold pieces fit together.

By Krish Iyer, strategy lead for automotive, & Said Tabet, distinguished engineer, Dell Technologies

Jigsaw puzzle junkies know, getting to the final triumphant piece takes time and patience. Similarly, it’s a long road from here to building driverless vehicles for large swathes of the population. A lot of pieces need to interlock first.

This is why the past few years have been frustrating, with forecasts made and missed. For some, the “the bloom is off the rose”.

But autonomous mobility hasn’t been consigned to the fantastical heap yet, and most probably won’t, given the steps the industry has taken to develop more realistic expectations of what can be achieved in a given timeframe–along with a greater understanding of the incremental feats that need to be pulled off first.

Hence, billions of dollars are still being poured into research and development to create what is, in essence, a highly sophisticated, data-centric, distributed next-generation compute infrastructure infused with Artificial Intelligence algorithms, and communication capabilities for responding to events in real-time, taking dynamic, ever-changing contexts into consideration at all times.

In fact, General Motors recently announced it’s investing $300 million in Chinese autonomous driving startup Momenta to develop mass-production vehicles with self-driving functions to gather real-time data. Globally, GM has earmarked $35 billion through 2025 for electric vehicles and autonomous driving technologies. While these figures point to the scale of momentum and excitement around autonomous vehicles, they also signal the complexity of the challenge.

A lot of money is needed because the task at hand is so darn hard. The industry is discovering (rather painfully) that mass-producing autonomous vehicles is a huge jigsaw puzzle. From a technology standpoint alone, it’s not always clear which pieces connect where. The hope is, players won’t teeter on the edge of completing this leviathan and complex innovation (after an awful lot of striving), only to realize they’re missing vital capabilities.

So, regulations and policies aside (which are desperately lacking), which interconnecting pieces are critical components for future driverless mobility? Aligning the following six technologies will most certainly be a prerequisite.

  1. New generation of processes
    Driverless mobility requires a hybrid AI model (i.e. both statistics-based and context-aware ML algorithms) that can analyze data generated by computer vision algorithms in real-time. However, AI models today are typically trained on graphical processor units (GPUs) running in the cloud. Installing GPUs is cost-prohibitive and requires constant power that no vehicle can afford to have, all the time. Hence, work is underway on a new generation of processors optimized for inferencing and transfer learning on edge computing platforms. These processors will need to be tightly managed to allow for continuous learning and monitoring of drift in AI algorithms.
  2. Zone gateways and electronic control units (ECUs)
    Another jigsaw piece is the realization that vehicles need to be outfitted with more powerful ECUs with greater memory, local storage and faster processing capabilities. This is needed to support the latest crop of advanced driver assistance systems (ADAS), that can collect data over 5G wireless networks to be analyzed in the cloud using machine learning algorithms. Additionally, consolidating ECUs into a domain controller running virtual infrastructure can help eliminate ECU sprawl.
  3. Fail-safe ADAS
    Meanwhile, ADAS platforms will need to run continuous updates as more data about driving conditions, such as the location of potholes and the impact of inclement weather, becomes available. And in a fully autonomous vehicle, they’ll need to be able to switch seamlessly to a failsafe model upon failure. With human intervention out of the picture, these systems will need to be designed with a high level of redundancy built-in.
  4. Edge-native architectures
    In the pursuit of continuous integration and service delivery capabilities, vehicle manufacturers are also chasing down a new generation of software-based on microservices that will make it simpler to manage applications running on ECUs via OTA updates. These architectures aren’t off-the-shelf operating systems. They’re specific to the automotive industry and its application development needs.
  5. 5G/6G architectures
    At the heart of autonomous cars is machine-to-machine communications, which requires 5G/6G as the underlying connective fabric. However, building a dense 5G infrastructure is an immense undertaking. Scaling to hundreds of thousands of cell sites alone takes time–and is incredibly complex. To achieve this, we need to pioneer a new way of approaching an infrastructure roll-out, that harnesses the speed of the cloud and IT worlds (using approaches like virtualization, containerization, APIs and service mesh) to do incremental work much faster. Of course, anything new requires new workflows–which perpetuates even more jigsaw pieces.
  6. Infotainment
    The future of driverless mobility will require holistic human-machine dialog and immersive experiences integrating 3-D visualization and AI. That’s quite a leap forward from today (which just displays relevant vehicular information and casual entertainment) and will need to happen in tandem with all of the above, and in concert with third-party infotainment providers.

Motoring along

In a Dell Technologies survey, 87% of IT decision-makers predicted that within 3-5 years autonomous vehicles “will be safe because they’ll constantly be learning the rules of the road and how to respond to unpredictable behavior”. This is undoubtedly over-optimistic, given all that needs to be overcome first (technology-wise, as well as policy-wise-, for instance, we still don’t have legislation that clarifies culpability in the event of an accident involving an autonomous car) but we are inching closer to this vision.

Chances are, we’ll be able to road-test autonomous mobility in the logistics industry before we roll out consumer vehicles (given the chronic shortage of drivers and more predictable settings, such as well-defined highways, truck stops).

Either way, a massive digitization effort is underway, as shown in a survey of 318 global automotive industry technology professionals conducted by Dell Technologies and Wards Auto.

Between now and 2025, nearly half of respondents (45%) plan to deploy a new in-vehicle architecture. And slightly more (46%) say those platforms will be running some type of software based on a service-oriented architecture such a microservices in the same timeframe.

The cumulative outcome of these endeavors won’t hit the road for another decade at least. But with all 1000-piece puzzles, contenders need to remain focused on the detail, while striving towards the final picture.

AI combined with next-generation ECUs, running software interconnected with cloud services over a high-speed 5G wireless network will one day transform both the driving and entertainment experience in our vehicles.