In the automotive industry major players already have their own simulation implementations tailored to their specific needs. However, they are constantly searching for tools that they can reuse their content in and connect their modules to, or that provide additional levels of validation and verification. This is why aiSim is built in a modular architecture connected by simple interfaces, making it easy to integrate into existing pipelines. As a result, AImotive's partners can rely on their in-house physics and sensor models while building on the aiSim engine. Reversely, our partners can integrate a single module of aiSim into their existing simulator. Due to the operating system and hardware agnostic nature of aiSim the modules can be used without changes to hardware infrastructures already in place.
Task scheduling solutions built into aiSim support the optimal utilization of multi-CPU and multi-GPU platforms. However, the simulator is completely hardware agnostic and can run with or without visualization on any hardware platform. The core of aiSim is our purpose-built engine. The aiSim engine ensures deterministic running and physics based-rendering building on the Vulkan API. Physics-based rendering ensures a higher degree of correlation between virtual and real-world runs, while determinism guarantees the repeatability of tests. Due to aiSim’s modular design our partners can approach the simulator as an SDK and implement its different modules while continuing to enjoy its core advantages.
Over 100 km of real-world roads have been modeled in aiSim. Highway sections from Sweden, France, Hungary and California, urban areas of Las Vegas, Budapest and El Camino Real in California are provided to our partners. Existing locations can be supplemented through the semi-automated content creation pipeline. Data from a wide range of input sources including photogrammetry, LiDAR point clouds, video footage, HD maps can be imported to aiSim. The content library also contains a wide variety of traffic agents including over 50 types of vehicles, articulated pedestrians and animals. To add further variability a wide range of weather patterns and other environmental factors can be altered in simulations.
aiSim’s standalone Scenario Editor enables the easy creation of scenarios for testing automated driving systems. To ensure efficiency scenarios can also be defined in a custom procedural language. Once the goal, expected behavior and constraints of a scenario have been specified, thousands of variants can be generated based on variable elements of the environment. AImotive offers a set of over 1000 scenarios defined by our Functional Safety Team based on functional safety engineering and NHTSA and Euro NCAP databases. To ease integration into existing pipelines, scenarios can also be imported through open formats, such as OpenDrive and OpenScenario.