Scenario generation for autonomous vehicle simulation
By Mike Brezonick22 January 2021
dSpace and its group company understand.ai, a specialist in automated, AI-based data annotation, are offering a new service that generates simulation scenarios from recorded measurement data used to validate functions for autonomous driving and driver assistance systems. With this offer, dSpace said it is supporting its customers in the heavy-duty vehicle and automotive industries in developing autonomous vehicles more efficiently using realistic simulation.
To reliably validate autonomous or semi-autonomous vehicles, thousands of near-realistic scenarios, including rare events, are required. Manually creating the rare events in special editors is extremely time-consuming. “With the Scenario Generation Service, we bring the complexity of the real world into simulation and enable validation with thousands of relevant and critical simulation scenarios,” said Product Manager Thorsten Püschl.
The Scenario Generation Service from understand.ai and dSpace uses existing sets of data recorded during measurement runs. In a highly automated process, AI-based annotation solutions from understand.ai are used to extract the relevant information from the raw data of the vehicle sensors. This creates realistic and consistent simulation scenarios. Optionally, data from object lists can also be used for scenario generation.
The generated scenarios are used to create exact reproductions of real operating situations in the simulation that help simulate events from test drives in the laboratory or compare simulations of sensor models with measurement data for sensor model validation. Additionally, detailed 3D models of the vehicle environment can be generated for physical sensor simulation.
The generated scenarios can be immediately used in the dSpace ASM simulation environment as well as the in existing dSpace infrastructure for software-in-the-loop (SIL) and hardware-in-the-loop (HIL) testing across multiple development stages.