Abstracting Reality to Ensure the Safety of Autonomous Driving

Try the DEMO!
Banner image
feature image

Our Way of Abstracting Reality

automated from concepts and real-world data to scenarios for simulation to test automated driving functions

  • use our or your data
  • automatic processing
  • analyze and compare on the dashboard
  • search easily for complex combinations
  • get concrete OpenSCENARIO/DRIVE files
Try the DEMO!
feature image

Input

One click to input your data!

  • automatic processing
  • private data or on-premise solutions
  • data compatibilitity through OMEGA-format
  • request any LevelXData dataset
  • LevelXData available (inD dataset)
  • converters for Lanelet2, OpenDrive, …
Try the DEMO!
feature image

Analysis

Get to know your scenarios!

  • compare number of scenarios
  • get scenario parameter distribution
  • get a sample out of a logical scenario
  • compare data sources
Try the DEMO!
feature image

Selection

Get interesting scenarios!

  • search for relevant scenarios
  • get concrete scenarios from distribution
  • visualize scenario
Try the DEMO!
feature image

Simulation

Export for simulation!

  • sampled from logical scenarios
  • behavior encoded in actions
  • get concrete OpenSCENARIO and OpeDRIVE files for test cases
  • tested with ESMINI
  • CARLA OpenX enhancements under development
  • experience in Highly Dynamic Driving Simulator
Try the DEMO!

Concepts

About the Concepts

For efficient structuring, storing and handling, an underlying scenario concept is developed for the database. It follows the categoriation of the 6 layer model and describes scenarios relatively to an ego road user. Thereby, scenarios are described and modeled on different abstraction layers to suit the users accordingly to its use case. A scenario is thereby defined as a sequence of scenes which can include actions and triggers and has an assigned ego road user. A scenario has a duration of typically 10 seconds and includes e.g. an intersection, a roundabout or a usual street.

Levels of Abstraction

For systematic description and easy usage, scenarios are described on different abstraction layers.

  • Functional/ abstract scenarios: On the highest level scenarios are described verbally for easy searches.
  • Logical scenario classes: Derived from abstract scenarios, logical scenario classes serves as a blueprint for parameters, attributes and relations to detail the abstract scenario and to generate concrete scenarios.
  • Logical scenario instances: Filled with actual data, distributions within a logical scenario class are modeled.
  • Concrete scenarios: Sampling logical scenario classes, concrete scenarios arise and can be used for simulations in OpenX format.

Attributes and Parameters

Scenarios are described by attributes and parameters for optimized seach and generation.

  • Attributes are used to describe scenarios comprehensively including positioning, timetoX metrics (e.g. THW, TTC, PET, timegap), occlusion, events and many more. So, for scenario can be searched via multiple attributes according to the actual use case. So, attributes aim to make every important aspect of a scenario searchable.
  • Parameters are dedicated to describe a scenario for generation. Inspite of attributes, parameters are thereby distinct or relations a re made explicit to prevent meaningless constellations and parametrizations. Thereby, different parametrizations are possible to describe the same scenario according to the actual needs.

Base and Focus Scenarios

How much complexity is needed? We describe all traffic relations between ego and object in any structured environment in less than 300 base scenarios. Easy to search - easy to generate. Thereby, those scenarios are build on a comprehensive ontology, so that scenarios used are derived systematically from given concepts. To allow for more complexity, combinations of those are called focus scenarios and can be used for more complex modeling. Within focus scenarios, multiple base scenarios are combined sequentially, in parallel or are adapted from the original base scenario to serve the dedicated use case.

Replay and Advanced Replay to Sim

Next to parametrizations of scenarios which abstract the reality, highest accuracy to simulate scenarios is done in a seperate branch without parameters for abstraction with replays of real-world data. Thereby, all components behave as they did in real-world. To still account for flexibility of scenarios according to potential ego behaviors, an advanced replay to sim is developed. This accounts for changes and adapts it to improve the realisticness of the new situation in OpenX standard.

Coverage and Completeness

How many scenarios are enough for sufficient testing? And is a proper set of scenarios chosen? For this, a method for proving the completeness of the scenario concept is used and applied to our scenario concept. Furthermore, coverage based on real-world data in the database can be approximated based on the limitations of input data.

References

Our scenario concept was developed and aligned in the safety assurance research project VVMethods . Learn more in our publications:

The Team

Christoph Glasmacher

Christoph Glasmacher

Research Scientist at ika

Michael Schuldes

Michael Schuldes

Research Scientist at ika

Timo Woopen

Timo Woopen

Manager Research Area Vehicle Intelligence & Automated Driving at ika

Univ.-Prof. Dr.-Ing. Lutz Eckstein

Univ.-Prof. Dr.-Ing. Lutz Eckstein

Director of the Institute for Automotive Engineering (ika) & President of VDI

Contact