How artificial intelligence and analytics will evolve autonomous driving and next-generation cars

Building autonomous cars takes more data than automotive research engineers ever imagined possible. To put some numbers around the autonomous driving issue, some current estimates predict that it will take around 23 to 25 million kilometers of driving to collect data and train the neural networks that will make autonomous cars feasible. Evolving autonomous driving could generate in excess of 100 exabytes of data and would take 1 million person years per simple capability to label the data.

Automakers are seeking simplified data access, analytics and training in flexible, scalable autonomous driving products that can manage and analyze data for each specific automotive use case. The technology needs to distribute data, analytics and training across multiple locations worldwide and enable fast development cycles based on an agile software development approach and fully automated inference and testing.

Key challenges facing automakers

Consumers need to understand that autonomous cars are still in the build phase. It will take extensive testing to get autonomous systems to drive a car like a human being and to be accepted by the various regulatory authorities. These new autonomous cars have to stop at a stop sign 100 percent of the time the way a human would. And they have to step on the brakes every time a pedestrian crosses the street unexpectedly. Anything less is unacceptable.

Creating the driving technology environment for evolving autonomous cars requires automakers to face three key challenges they’re asking themselves behind the scenes:

How long will it take to meet our goals?

The Brookings Institution reported that automakers have invested some $80 billion in autonomous vehicles over the last three years. Making autonomous cars a reality will require a long-term commitment that won’t be accomplished by the end of 2018

SAE International published a list of six automation levels (0-5) that cars must go through before they reach full automation. To give you an example of the timeframe, automakers should expect it to take at least one year to fully establish Level 3, when cars are primarily automated but the system asks the human to intervene in certain situations.

It could then take another two to three years to reach Level 4, in which the car safely runs autonomously, primarily in many driving modes and situations.

And it will take nearly five years to reach Level 5, or fully autonomous driving with no human intervention.  This is the hardest level to fulfill.

What kinds of special capabilities will autonomous cars require?

Overall, SAE International specifies 10 to 12 different car system capabilities to achieve Level 5, but three are the most critical:

  • Object recognition and tracking: The ability to identify and track objects on the road — be they pedestrians, signs, buildings or traffic lights — like humans do through their own perception.
  • A situation analyzer: The ability to compute and analyze gigabytes of data per second in real time to deliver data features that enable the motion predictor system to make intelligent decisions on the road.
  • A motion predictor: The ability for artificial intelligence to handle critical situations that come up while driving, for example a pedestrian running across the street, a car being sideswiped or hazardous weather conditions.

What kind of back-end computing systems will we need?

The sheer volume of data required by autonomous cars is unprecedented and — without smart optimization — will require exabytes of storage with enormous bandwidth capabilities. In fact, in 2017 alone, the entire world produced around 120 exabytes of data. These data volumes are too large be managed by one company.

Given the volume of data needed for autonomous driving, it would be impossible to transfer all this data across an enterprise network. In the past, data was only stored in large data centers. Today it’s possible to store large amounts of data in smaller decentralized data centers all over the world. Realistically, it makes more sense to keep the data where it occurs and share the data wrangling programs, analytics algorithms and machine learning models with everyone across the organization based on a federated approach.

By storing the data, programs and driving model knowledge by federation and then leveraging on-premise and cloud environments, automakers can develop a platform in which their teams around the world can develop autonomous driving capabilities wherever they are.

A safe and profitable future for autonomous cars

Automakers must change from a deterministic operations research approach based on driving business rules and current programming models to an environment that’s driven by artificial intelligence, big data and analytics, further-developed sensors and computer science protocols.

Clearly, fully autonomous cars are five to seven years into the future. But based on the multi-billion dollar investment the automakers have made, we believe it’s possible for the industry to create a future for autonomous driving that’s both safe for consumers and profitable for the industry and the global economy.

Davor-Andric-headshot-loresDavor Andric is CTO of AI & Analytics for DXC North and Central Europe. Over the last 20 years, Davor has been working in the consulting, software and technology space. His expertise is in designing and building scalable platforms for analytics, machine learning and AI, and building products on those platforms.

Speak Your Mind


This site uses Akismet to reduce spam. Learn how your comment data is processed.