Today AI has disrupted many of the traditional manufacturing industries, with some of the fastest changes taking place within automotive, where GPU-accelerated technologies have replaced century-old working practices. Right now, every aspect of the business is being transformed, from logistics and supply chain management, to design and testing, and sales and after-service. Moreover, as autonomous vehicles complete their journey from R&D labs and onto public roads, AI is at the core of the new generation of services and will be pivotal to the industry’s future.
While cars still play the traditional role of getting consumers from A to B, they remain a signifier of lifestyle and status, and are transforming into sophisticated digital products where the software is equally as important as the hardware. Most of this new AI-based functionality requires high performance computing (HPC) infrastructure for machine learning, and highly attuned parallel processing that is hard to deliver using traditional CPU-focused architectures.
The car industry’s demand for AI and ML hasn’t been lost on NVIDIA, for example, who has launched several innovative solutions aimed at the automotive sector. They include its DRIVE Hyperion development platform, DRIVE AGX Orin on-board computer, DRIVE Sim software, and DRIVE Constellation – a hardware platform based on NVIDIA’s DGX servers, designed to run car-related workloads in a high performance data centre environment.
These technologies are driving major advancements within the sector, and British manufacturer Jaguar Land Rover recently became one of the first luxury car brands to utilise AI-powered compute to transform its product offering with NVIDIA. Together they are jointly developing autonomous driving and connected services for all future vehicles using NVIDIA DRIVE, leveraging in-house developed data centre solutions with NVIDIA DGX™ for training advanced AI models, and using DRIVE Sim software built on NVIDIA Omniverse™ for real-time accurate simulation.
Design and test
Historically, the car design process relied on human expertise, intuition, and iteration, but this approach is fast-being replaced by software and mathematical models.
TÜV SÜD, the German corporation known for its vehicle crash test certification, offers a wide range of virtual testing services for autonomous vehicles as a means to discover any weak points in the design or its components under real-world conditions. This is prior to any prototype being produced. Later in the process, AI is also used on the production line, to detect defects and signs of botched assembly, and even track the amount of dust on the car body that could interfere with key consumer finishing aspects like the vehicle’s external sensors.
Machine learning systems are also proving invaluable for electric vehicle (EV) research and development (R&D), where it is used to improve power efficiency by analysing the impact of equipment like heated seats, air conditioning, entertainment systems, and more. This approach enables EVs to drive for longer on a single charge and helps to address one of the key challenges of EV adoption, compared to fossil-fuel powered vehicles – charging!
AI is also changing the ownership experience of supercars, as more high performance marques launch EVs where part of the familiarisation process for new owners, is accepting the concept of 'let the car decide', in particular, managing range and power.
Through a continuous learning process, the car’s AI system builds a profile of the users driving habits and its rate of power consumption. Some EVs also incorporate a sophisticated management system with ultra-fast charging capabilities. This feature allows the car to pre-condition and prepare the battery to charge at the fastest rate possible.
Living in a simulation
Autonomous vehicles and self-driving cars are not a new concept. The first experiments with algorithm-controlled vehicles took place back in the 1980’s, but it took another 40 years to refine sensors and computer vision to a point where they could guide cars along any road, and not just carefully mapped, empty circuits.
For many, the initial years of commercial autonomous car development presented a ‘chicken or egg’ scenario where the machine learning models used to control these systems needed be made safe through extensive testing. Manufacturers were, of course, unable to test them on public roads until they were deemed sufficiently safe.
The answer to this conundrum, as is often the case, came from an unusual and innovative direction - why not create expansive virtual worlds that would replicate all the key characteristics of real roads?
To cater for virtual automotive testing applications, NVIDIA created ‘NVIDIA DRIVE Sim’ as part of its Omniverse digital twin platform. The software is intended for autonomous vehicle testing in a wide range of scenarios, and offers groundbreaking render qualities that could, almost, be mistaken for real-life video footage.
This approach allows autonomous vehicles to drive millions of virtual miles, without any danger to drivers or pedestrians, and countless self-driving car technology developers have embraced this kind of virtual testing as part of their R&D.
They include Alphabet-owned Waymo, Tesla and UK AI Start-up, Wayve who has become a key name in the autonomous vehicle race. Wayve it also incredibly interesting, as it chooses to utilise deep learning and cameras rather than opt for Light Detection and Ranging (LiDAR) technology to guide its vehicles.
AI changing the sector – inside and out
Today AI is changing not just how we drive, but how we interact with the car and how it, in turn, interacts with us. A simple example is virtual digital assistants (VDIs), which have become commonplace within our homes, and in some car models.
Another exciting application of AI is in-car sensing, whereby manufacturers will equip the car with driver and passenger monitoring. This can be used to automate aspects of the in-car experience like lights or music, but it can also improve safety features. For example, new systems are being developed to detect signs of drowsiness, so if the driver keeps his eyes closed, the car will quickly realise they are likely asleep, slow down the vehicle, and park it on the side of the road.
What’s clear is that as car technologies evolve, AI will become crucial to both manufacturing, and the consumer experience. As autonomous vehicles become a reality, the ability to store, process and analyse data within a secure and localised, high performance data centre will be critical for both safety and innovation, and to give automotive manufacturers a competitive edge.
----------
The full and edited version of this article was recently published on AI Business, read it here.
Lee Myall
Lee Myall is the former CEO at Kao Data and has spent more than 20 years leading teams that design, develop and deliver mission-critical digital infrastructure.