Intersection of AI and Simulation in the Automotive Industry

The automotive field has historically been a rich area of innovation, with increasing vehicle complexity and tight production schedules requiring the adoption of new tools and techniques to build a differentiated product. More recently, automotive engineers are faced with new obstacles as they are tasked with integrating artificial intelligence (AI) into vehicle systems and R&D workflows. AI cannot be viewed as a standalone component, but must be integrated with consideration for the impact on adjacent systems, leading to a requirement that AI algorithms are simulated alongside other components to understand their impact and function before they can be deployed to the vehicle.

At a high level, there are three key points of interaction between AI and simulation in the automotive industry. The first has to do with addressing the challenge of insufficient data, as simulation models can be used to synthesize data that might be difficult or expensive to collect. The second is the use of AI models as approximations for complex high-fidelity simulations that are computationally expensive, also referred to as reduced-order modeling. The third is the use of AI models in embedded systems for applications such as controls, signal processing and embedded vision, where simulation has become a key part of the design process.

As automotive engineers are finding new ways to develop more effective AI models, this piece will explore how simulation and AI combine to solve challenges of time, model reliability and data quality.

Challenge 1: Data for Training and Validating AI Models

The process of collecting real-world data and creating good, clean, and cataloged data is difficult and time-consuming, particularly in the automotive field. Engineers also must be mindful of the fact that while most AI models are static (they run using fixed parameter values), they are constantly exposed to new data and that data might not necessarily be captured in the training set.

Projects are more likely to fail without robust data to help train a model, making data preparation a crucial step in the AI workflow. ‘Bad’ data can leave an automotive engineer spending hours trying to determine why the model is not working, without the promise of insightful results.

Simulation can help automotive engineers overcome these challenges. In recent years, data-centric AI has brought the AI community’s focus to the importance of training data. Rather than spending all a project’s time tweaking the AI model’s architecture and parameters, it has been shown that time spent improving the training data can often yield larger improvements in accuracy. The use of simulation to augment existing training data has multiple benefits:

• Computational simulation is in general much less costly than physical on-vehicle experiments.
• The automotive engineer has full control over the environment and can simulate scenarios that are difficult or too dangerous to create in the real world—such as emergency braking on icy roads or near-miss collision navigation with autonomous vehicles.
• Simulation gives access to internal states that might not be measured in an experimental setup, which can be very useful when debugging why an AI model doesn’t perform well in certain situations—including testing the viability of models predicting nonlinear values like NOx (nitric oxide and nitrogen dioxide) emissions.

With a model’s performance dependent on the quality of the data it is being trained with, automotive engineers can improve outcomes with an iterative process of simulating data, updating an AI model, observing what conditions it cannot predict well, and collecting more simulated data for those conditions.

Using industry tools such as Simulink and Simscape, automotive engineers can generate simulated data that mirrors real-world scenarios. The combination of Simulink and MATLAB enables engineers to simulate data in the same environment that they build their AI models, meaning they can automate more of the process and not have to worry about switching toolchains.

Challenge 2: Approximating Complex Systems with AI

When designing algorithms that interact with physical systems, such as an algorithm to control a hydraulic valve, simulation-based modeling of the system is key to enabling rapid design iteration for your algorithms. In the controls field, this is called the “plant model;” in wireless vehicle communications, it is called “channel model.” In the reinforcement learning field, it’s called the “environment model.” Whatever you call it, the idea is the same: create a simulation-based model that gives you the necessary accuracy to recreate the physical system your algorithms interact with.

The problem with this approach is that to achieve the “necessary accuracy” automotive engineers have historically built high-fidelity models from first principles, which for complex systems can take a long time to build and simulate. Long-running simulations mean that less design iteration will be possible, resulting in not enough time to evaluate potentially better design alternatives.

AI comes into the picture here in that automotive engineers can take that high-fidelity model of the physical system that they’ve built and approximate it with an AI model (a reduced-order model). In other situations, they might just train the AI model from experimental data, completely bypassing the creation of a physics-based model. The benefit is that the reduced-order model is much less computationally expensive than the first-principles model, meaning that the automotive engineer can perform more exploration of the design space. And, if a physics-based model of the system does exist, the automotive engineer can always use that model later in the process to validate the design determined using the AI model. It doesn’t need to be the entire system that is approximated; the user can decide and prioritize which parts of the system are best-suitable for AI approximation by assessing the speed and accuracy trade-off.

Another benefit of using AI-based approaches for reduced-order modeling comes from reducing the number of software tools that need to be combined to simulate a system. For example, when trying to add detailed controls to a system or when assembling subsystems into larger models (often done in Simulink), reduced-order modeling can help reduce other software dependencies, functional mockup unit/functional mockup interface integration, and other hurdles associated with co-simulating various automotive engineering tools.

Recent advances in the AI space, such as neural ordinary differential equations, combine AI training techniques with models that have physics-based principles embedded within them. Such models can be useful when there are certain aspects of the physical system that the automotive engineer wishes to retain, while approximating the rest of the system with a more data-centric approach.

Challenge 3: AI for Algorithm Development

Automotive engineers in applications like control systems have come to rely more-and-more on simulations when designing their algorithms. In many cases, these automotive engineers are developing virtual sensors, observers that attempt to calculate a value that isn’t directly measured from the available sensors. A variety of approaches are used including linear models and Kalman filters.

But the ability of these methods to capture the nonlinear behavior present in many real-world systems is limited, so automotive engineers are turning to AI-based approaches, which have the flexibility to model the complexities. They use data (either measured or simulated) to train an AI model that can predict the unobserved state from the observed states, and then integrate that AI model with the system.

In this case, the AI model is included as part of the controls algorithm that ends up on the electronic control unit, which has performance/memory limitations, and typically needs to be programmed in a lower-level language like C. These requirements can impose restrictions on the types of machine learning models that are appropriate for such applications, so the automotive engineers may need to try multiple models and compare trade-offs in accuracy and on-vehicle performance.

At the forefront of research in this area, reinforcement learning takes this approach a step further. Rather than learning just the estimator, reinforcement learning learns the entire control strategy. This has shown to be a powerful technique in some challenging applications such as robotics and autonomous systems but building such a model requires an accurate model of the environment, which may not be readily available, as well as the computational power to run a large number of simulations.

In addition to virtual sensors and reinforcement learning, AI algorithms are increasingly used in embedded vision, audio and signal processing, and wireless applications. For example, in a car with automated driving capabilities, an AI algorithm can detect lane markings on the road to help keep the car centered in the lane. In an in-vehicle voice assistant, AI algorithms can help enhance speech and suppress noise. Simulation is used for integration testing to ensure the overall design meets requirements.

The Future of AI for Simulation

In the automotive industry, as models grow in size and complexity, AI and simulation will become even more essential tools in the engineer’s toolbox. Industry tools like Simulink and MATLAB have empowered engineers to optimize their workflows and cut their development time by incorporating techniques such as synthetic data generation, reduced-order modeling, and embedded AI algorithms for controls, signal processing, embedded vision and wireless applications.

With the ability to develop, test and validate models in an accurate and affordable way before hardware is introduced, these methodologies will only continue to grow in use.

Seth DeLand is Data Analytics Product Marketing Manager at MathWorks.