Variations in the parameters leads to the new scenarios which are in tens of thousands. Among the all scenarios engineers need to be smart to address the most relevant scenarios with a credible range of parameters to model the edge cases. They must envision scenarios that ADAS might encounter to ensure a broad range of parameters that would capture edge cases, where the systems are the most challenged. Edge cases are problems or situations that occurs only at an extreme (maximum or minimum) of the operating parameters. The relevance of a scenario and the edge cases changes with the change is the design of the ADAS / AV, so considering the fix set of scenarios with fixed parameters to test is not enough. VTD Scale has a flexible framework which can apply parameters to both the simulation model and scenario. Parameters can be directly mapped to fields in any model-data file (for both environment and System Under Test) and generative applications, which parametrically create on-the-fly simulation model files. This permits parametric control of static content (such as road decoration) and dynamic content (such as placement of pedestrians, obstacles, etc.).
Cloud-based Scalability and CPU/GPU-based Simulation
Smart Autonomous vehicles create unique set of opportunities and challenges. In order to test drive billions and billions of virtual miles through computer simulations, cloud-based computing services are required. These cloud-based data centers enable thousands of CPUs and GPUs to perform 100,000’s of parallel simulations to build, test, deploy, and manage these complex simulations. With the full scalable support of the both CPU and GPU-based Simulations, VTD Scale utilizes standards to enable large-scale simulation, in which 1000’s of containerized simulation flows based on the VTD software are executed in parallel.
Full Integration with VTD/ Containerized Services
Among the complex network of different systems and functions in any smart car, the essential components boil down to software’s and sensors. With an increasing complexity in the interactions between the different systems in the vehicle such as sensory (optical, Lidar, radar, ultrasound sensors and the driver/controller logic), environment (light, weather, obstacles, pedestrians, road and traffic conditions), and the physical behavior of the vehicle (structural, crash, noise, vibration, handing, ride, comfort, durability, fatigue), the requirement for testing and validation is increasing exponentially. VTD Scale has a flexible framework to link together VTD with other containerized services (sidecars); which include the sensor, controller and physical vehicle simulators. The full richness of the VTD communication mechanisms (such as FMI) is supported. VTD Scale can manage different versions and model fidelities of the components. During the VTD Scale studies, the components of the simulation can dynamically be compiled to the setup for a co-simulation.
All Road Networks Usable
VTD Scale offers an opportunity to use the road networks that are developed based on one’s requirement with the Road Designer or can you use the large geo-specific road networks that are either developed through the Road Designer or are extracted from the mapping systems like Leica Geosystems and resembles to original locations.
Scripting and Automation
VTD Scale utilizes Python to allow the creation of any study sequence (template), allowing the integration of open and closed-loop sampling methodology. VTD Scale supports more advanced approaches beyond the traditional (single step) Design of Experiments (DOE) or stochastic sampling strategies. This allows the inclusion of Artificial Intelligence or Machine Learning based methodologies to identify and focus on edge case subspaces (sub-populations) during the study.
In the execution of thousands of parallel simulations, large amount of data is generated which on processing by record by record doesn’t address the solution to the analysis of simulations for edge case detection. With the more advanced real-time analysis, VTD Scale allows to directly create and compress the data such as KPIs and annotations generated during the parallel execution of simulations. It is crucial to keep the result data volume in bounds and prepare the data for immediate access to data analytics tool. This enable direct creation and training data for any type of supervised or unsupervised training.
Integration to Data Analytics Tools
The well-organized data helps the engineers and data scientist to run and validate the parametric studies. VTD Scale presents results in standard data structures which through the use of data adaptors, offers connection to all data analytics tools. This allows engineers and data scientists to directly process and analyze the study results. The precise data assist the engineers to understand the key performance metrics of ADAS / AV functions while for the data scientist it is useful for reinforcement learning and in Artificial Intelligence pipelines.