Individual Virtual Environment and Sensor Simulation – iVESS

It takes sophisticated image synthesis capabilities, richly detailed textures and geometric shapes, and realistic lighting models to map complex real-world situations and environments to the virtual realm. Computer games have much the same requirements. A satisfying gaming experience demands photo-realistic computer graphics, fast frames-per-second rates, and efficient use of hardware and its parallel architecture. Computer graphics techniques such as ray tracing enable consistent, synthetic, multi-sensor data generation in real time. With this method, we can simulate various types of sensors such as ultrasound, radar, and cameras. And that lets us do many things such as calculate physical effects in real-time, generate test values for disruptive factors in reality, and create ghost targets.

Autonomous systems use different types of sensors – cameras, LiDAR, radar, and ultrasound – to scan their surroundings.

This is why virtual sensors operated in a richly detailed simulated world have to deliver the same data quantity and quality as their real-world counterparts. Various sensor simulation models and interfaces can be applied to suit the required level of testing, which is determined by how deeply the control system is integrated.

The finite element method (FEM) numerically simulates a sensor’s physical properties with great precision, but requires a lot of computing time and is often limited to static environments. Simple bounding box methods can be executed in multiples of real time, but their representation of the virtual environment is also greatly simplified. With these drawbacks, neither option is particularly well suited for verification and validation.

A new way of tackling this problem is ray tracing, which efficiently computes algorithms on graphics cards (GPUs) to generate accurate and highly detailed data for dynamic scenarios. Sensor data can thus be simulated for each of the aforementioned sensor types. This data then serves as input for downstream processing algorithms and sensor data fusion. It is also possible to have the simulation deliver labeled sensor data in an automated routine. This information can then be used to train and test AI-based systems.

Is your existing simulation environment missing a key feature in 3D animation? Do you want to simulate sensors at specific positions that your current tool does not allow? Are you looking for a solution that extends your existing environment without having to rebuild the entire system? Do you want to become independent of specific suppliers and take an open approach to future developments? Do you want to avoid high license costs, especially when simulating multiple instances?

With GPUs running in parallel, the simulation generates raw data as fast as or even faster than the real system’s sensors. This enables efficient closed-loop simulation with actual ECUs. Our iVESS framework combines game engine, ray tracing, and physics models. It can simulate any 3D environmental scenarios for autonomous systems as well as the data furnished by their sensors. The application can run as a co-simulation with third-party software and in standalone mode. Equipped with a range of input and output interfaces, it is highly compatible with legacy data formats and toolchains.

Take a look at our project examples.

Our iVESS framework lets you realize any scenario. Rather than taking a product-based approach, we tailor the solution to readily integrate into the workflow you have in place now. In other words, we integrate the solution into your toolchain or hardware to suit your unique requirements. You get a custom simulation that fits. And if you wish, we will be happy to deliver it in form of a white-box solution. This way you can continue developing the solution as you see fit. You white-label it – that is, offer it to your customers under your brand. The iVESS framework adapts to your needs. Anything goes, from simulations and HiL applications to classic test scenario verification and validation.