How can the Virtual world save the Physical world from its growing complexity? Lessons learnt from the use of digital twin for Autonomous Driving

Autonomous driving is steering the future of automotive, and as such, so is the use of digital twins

The auto industry is gradually gravitating towards autonomous vehicles, and today, every modern car comes with features of advanced driver-assistance, which are all software-enabled. To allow carmakers to cover an infinite number of testing scenarios, Digital twin technology is now a must have. New insights can be continuously fed back onto edge servers, adapting vehicle software and optimizing decision-making on the road.

For any connected product becoming software-enabled, a digital twin is necessary to manage the complexity of its configurations and drive the multiple validation and certification test cases.

Complexity as a new standard

Truly autonomous vehicles have long been awaited for, as they are supposed to be safer, more reliable, and provide a much better user experience, removing the need for human involvement.?

But the path to genuine autonomy is still a long way ahead, with even industry leaders recognizing the gap.?The more advanced systems, like Tesla’s Autopilot or General Motors’ Super Cruise, only fall into Level 2 of driving automation, meaning that the vehicle can control steering and accelerating / decelerating, but falls way short of self-driving as a person still needs to be able to take control of the car at any time from the driver’s seat. Level 5, the highest level of vehicle autonomy,?would mean?that the vehicle is able to perform all driving tasks under all conditions while zero human interaction or focus is required.?

Though obvious legal challenges that befall autonomous vehicles, like the burden of responsibility in case of an accident, vastly limit the ability of carmakers to deliver on the autonomous promise, the overarching problem is the complexity of such fully autonomous vehicles. To consider a Level 5 vehicle, an almost infinite number of scenarios need to be modelled, be it for software-to-software as well as software to mechanical interactions, in addition to the sheer amount of data that needs to be analyzed in real time.?

Fully self-driving vehicles are subject to the following variables:?

• Changing speed regulations across and within countries (between highways, freeways, roads…)?
• Multitude of signaling panels that need to be recognized (stop, traffic lights, roundabouts, repair…)?
• Variety of weather conditions (to steer acceleration and deceleration)?
• Heterogeneous traffic and road conditions (feeding onto surrounding vehicles’ data to adapt its speed and direction in case of heavy traffic)?
• Evolving road architecture (new routes are being destroyed and rebuilt every day)?
• Potential hazards…

Autonomous driving programming is also subject to ethical scenarios in which vehicles must choose the appropriate behavior in case of a potential accident involving another vehicle or pedestrian. These micro-decisions are taken every day by human drivers but need to be thoroughly scenarized to allow for sensible decision-making at the vehicle-level. This is where the digital twin of the vehicle comes into play: carmakers digitally replicate vehicles in a virtualized environment to account for an infinite number of scenarios and adapt individual vehicle software accordingly.

While it is true that the level of complexity required to run a fully autonomous car is extreme (modern cars run on 100+ million lines of code, compared to 14 million lines of code for a Boeing 787 Dreamliner), the shift towards data-driven, software-enabled, complex products is prevalent across all industries and throughout all industrial value chains.?

?

Connectivity as the backbone?

Market trends are simultaneously moving towards the following dimensions:??

• Complexity: Products are getting more intricate and personalized, with an infinite combination of hardware and software system features, ever so changing client expectations, naturally increasing the diversity of scenarios to handle.?
• Digital: Products are becoming more digitally intertwined, with most of the value coming from these software-enabled features. Advanced driver-assistance features or ADAS (blind spot monitor, forward collision warning, intelligent speed adaptation, etc.) are gradually becoming standards – so are embedded navigation system and functions helping the driver reduce operating costs. To that extent, new cars are as much software as they are hardware.?
• Criticality: Regulations and increased reliance on software puts pressure on quality and safety. Moving from 99% to 99.9% reliability and covering a few new edge cases costs a lot more than moving from 90% to 99%.?Offering ADAS features is relatively easy in today’s world, and it can cover most problematic use cases – but ADAS-enabled vehicles still require human steering. Achieving carmakers’ goal of true autonomous driving, will increase the number of use cases that need to be covered. Doing so will require a genuine technological prowess to reach the level of confidence required on software safety and quality.
• Lifecycle: Over-the-air updates are increasing product lifetime, improving performance and durability after launch, maintaining an accurate view on the state of the installed base.?The over-the-air update system is a new source of complexity since it must consider fleet diversity in addition to meeting compliance standards.

?

Higher emphasis on digitally enabled features, complexity, criticality, and longer lifespan is mainly enabled by the increasing number of sensors and actuators that are embedded onto every asset, allowing manufacturers to digitally replicate them whether they are:?

• Components?(components or parts that make up a device) – e.g., a pipe?
• Devices?(devices that perform a function and that are made up of a combination of components) – e.g., a pump, a turbine
• Systems?(collections of devices that transform inputs into outputs – also called processes) – e.g., a braking system, a car
• Systems of systems?(collections of systems) – e.g., a fleet, a city?

?

The data coming from those IoT sensors are:?

• Diverse:?ranging from physical characteristics such as speed, weight, temperature, stress, torque; to statistical components such as fuel efficiency; as well as business elements such as pricing, costing, etc.?
• Vast: increasing number of IoT sensors means increasing – almost exponentially – data available for analysis
• Dynamic:?compared to simplified simulations, real time data is subject to unpredicted changes (e.g., shifts in production, supply chain, inventory, etc.)
• Intricate:?all interactions within an industrial value chain (R&D, material sourcing, supply chain, manufacturing, marketing, sales, aftersales) can be influenced by a single evolving datapoint?

?

Closing the gap between physical and digital as the only way forward?

?A 2016 study by the National Highway Transportation Safety Administration found that human error accounted for anywhere between 94% to 96% of all auto accidents. On top of the environmental and efficiency benefits provided for by the autonomous car, the main driver for its adoption is clearly safety.

?To provide maximum safety for the driver, other pedestrians, and vehicles with which it interacts, the autonomous vehicle will be able to continuously update during runtime, accounting for more accurate behaviors in any scenario.

The simplification of product architecture, enabling the software platform with the maximum amount of features and continuous adaptation, is not the only competitive edge. The optimized conditions to perform AI within the car (the virtual twin being able to support large computations offboard & OTA updates), AI powered onboard and offboard capabilities (leveraging vehicle-to-vehicle and vehicle-to-infrastructure real-time interaction), and the multiplication of onboard computation power thanks to the virtualization of the ECU network are also decisive. Mastering those critical features will be a key component for carmakers’ digital transformation, and leveraging digital twins is a powerful enabler securing its success. ???

?

All there is to know about digital twins

A digital twin is a generic term used to describe a complete virtual replica of a physical asset, however complex it may be (from small primary components within a machine to a major city like Singapore). It comprehensively and dynamically mirrors the operating environment of real-life assets, the interactions of the systems that are embedded onto it, and capture live data coming from connected sensors and captors, providing insights for better decision-making at every level (from a factory to fleet operations).

?It can include, but is not limited to, combinations of the following data: physics-based and model data, analytical models and data, historians, transactional data, visual models. Digital twins collect data centrally for every entity of a factory or every system within a product or a set of products, and then makes that information available to various areas of the business for their specific applications through APIs.

?The digital twin effectively interfaces three integral components, which are data (coming from various functions), models (computational and analytic models to describe, diagnose, predict, and simulate the states and behaviors of the real-world assets and systems) and services (ranging from descriptive insights to prescriptive actions):

Authors:

Vanessa Lyon , BCG, Managing Director and Senior Partner?

? Mika?l Le Mou?llic , BCG, Managing Director and Partner

? ??Jean-Christophe LAISSY?? , BCG, Partner and Director

? Laurent Alt , BCG, Associate Director, Enterprise Agility & Systems Engineering

?Alexandre Toureh, BCG, Senior Associate

?Victoria Guérendel, BCG, Associate?