Research Project PEGASUS. - pegasusprojekt.de · Cross-Verification Assessment of AD –...
Transcript of Research Project PEGASUS. - pegasusprojekt.de · Cross-Verification Assessment of AD –...
Research Project PEGASUS. Test Case Variation and Execution.
2
3
Cross-Verification
Assessment of AD – Methodology.
4 © PEGASUS | Goals and work contents of PEGASUS
Simulation
FOT / NDS
Proving Ground
Scenarios
Scenarios
Scenarios
Extracting
Input
Input
PEGASUS Database
Linking KPIs and Test Instance to
Scenario File
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Cross-Verification
Assessment of AD – Methodology.
5 © PEGASUS | Goals and work contents of PEGASUS
Simulation
FOT / NDS
Proving Ground
Scenarios
Scenarios
Scenarios
Extracting
Input
Input
PEGASUS Database
Linking KPIs and Test Instance to
Scenario File
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Cross-Verification
Assessment of AD – Methodology.
6 © PEGASUS | Goals and work contents of PEGASUS
Simulation
FOT / NDS
Proving Ground
Scenarios
Scenarios
Scenarios
Extracting
Input
Input
PEGASUS Database
Linking KPIs and Test Instance to
Scenario File
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Cross-Verification
Assessment of AD – Methodology.
7 © PEGASUS | Goals and work contents of PEGASUS
Simulation
FOT / NDS
Proving Ground
Scenarios
Scenarios
Scenarios
Extracting
Input
Input
PEGASUS Database
Linking KPIs and Test Instance to
Scenario File
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Evaluation
Traces KPIs
Cross-Verification
Assessment of AD – Methodology.
8 © PEGASUS | Goals and work contents of PEGASUS
Simulation
FOT / NDS
Proving Ground
Scenarios
Scenarios
Scenarios
Extracting
Input
Input
PEGASUS Database
Linking KPIs and Test Instance to
Scenario File
1 2
Information Basis for AD Assessment.
9 © PEGASUS | Goals and work contents of PEGASUS
FOT / NDS
Accident Database
Public Testing Fields
Expert Knowledge
Laws/ Standards/ Guidelines
PEGASUS Database
Test Case Data:
- logical scenario spaces,
- parameter lists with distributions,
- known critical concrete scenarios labelled within the logical spaces,
- metrics / KPIs,
- pass/fail criteria.
Scenario data model enrichment with a-priori knowledge about reality:
A-priori probabilities and correlations.
1
Scenario File
Input KPIs
Traces
Evaluation
AD function connected
Dynamic Parameter Space
Logical Scenario File
Parameter Variation
New Scenario Files
Simulation Framework
AD function connected
Stochastic Variation – Methodology. 2
Scenario Data Model for exchangeable and reproducible Scenarios.
1
EGO
1
2
3
4
5
6
…011010…
EGO
Concrete Scenario:
• Highway, 3 lanes, speed limit 80 km/h, • EGO drives const. 80 km/h, • Vehicle 1 flashes and cuts in, • sunshine, dry road, etc.
1
What is varied? – Example.
1
2
3
4
5
Levels of PEGASUS Scenario Data Model.
6
EGO
1
What is varied? – Example.
Concrete Scenario:
• Highway, 3 lanes, speed limit 80 km/h, • EGO drives const. 80 km/h , • Vehicle 1 flashes and cuts in at different starting positions with different trajectories.
1
2
3
4
5
Levels of PEGASUS Scenario Data Model.
6
Concrete Scenario:
• Highway, 3 lanes, speed limit 80 km/h, • EGO drives const. 80 km/h, • Vehicle 1 flashes and cuts in, • new static elements.
What is varied? – Example.
1
2
3
4
5
Levels of PEGASUS Scenario Data Model.
6
EGO
1
Concrete Scenario:
• Highway, 3 lanes, speed limit 80 km/h, • EGO drives const. 80 km/h, • Vehicle 1 flashes and cuts in with 100 km/h, • lane closes.
What is varied? – Example.
EGO
1
1
2
3
4
5
Levels of PEGASUS Scenario Data Model.
6
Concrete Scenario:
• Highway, 3 lanes, speed limit 80 km/h, • EGO drives const. 80 km/h, • Vehicle 1 flashes and cuts in with 100 km/h, • snow.
What is varied? – Example.
EGO
1
1
2
3
4
5
Levels of PEGASUS Scenario Data Model.
6
Stochastic Variation – Goal.
17 © PEGASUS | Goals and work contents of PEGASUS
Main Goal:
Find all relevant critical parameter subsets within each logical scenario space, that are within the use-case of the AD function. The stochastic variation within the PEGASUS project has two tasks:
- Exploration: Find critical parameter sets in the sense of a certain safety metric in the defined subspace.
- Characterization: Goal is to deliver a description of the parameter subspace where a metric reports a safety violation.
Stochastic Variation – Exploration.
18 © PEGASUS | Goals and work contents of PEGASUS
PEGASUS focused on two approaches of parameter space exploration:
1. Explore known critical scenarios, also with local variation of parameters.
param1
param2
2. Explore a given abstract parameter subspace (logical scenario) with optimization methods in order to find one or more worst cases of the safety metrics.
Starting points for iterative search may be chosen as known critical parameter subsets from real traffic data.
Stochastic Variation – Characterization.
19 © PEGASUS | Goals and work contents of PEGASUS
1. Characterization means assessing the probability of safety violations for a logical scenario space with a given confidence measure.
2. Test result is a description of the critical parameter subspace where a metric reports a safety violation including an approximation of probability.
Next Processing Step – Risk Assessment.
20 © PEGASUS | Goals and work contents of PEGASUS
Test Results:
- concrete test results per critical scenario
or
- description of the parameter subspace where a metric reports a safety violation including an approximation of extension and probability.
Contact:
BMW Group
Dr. Mark Schiementz
+49 151 60135903
www.pegasusprojekt.de
21 © PEGASUS | Goals and work contents of PEGASUS
Resimulation Architecture 2
• Real measurement data of relevant scenarios.
system under test
Actuator
Simulation Model
Environment-/ Situation-analysis
Fusion
Localization
AD Core-Function
Planning
Driving Strategy
Control
longitudinal
lateral
lights
Open Simulation Interface (OSI) OpenScenario / OpenDRIVE
NDS-Format
actuator model
OSI
Einviron-
ment Represen-tation
OSI
Simulation Framework
Te
st R
esu
lt E
valu
ati
on
system under test (AD Function) sensor model environment model
Actuator
Simulation Model
Vehicle-Simulation Model
USS
Map
Synth. Backend
Radar
Sensor Model
Lidar
Lidar Sensor Model
Camera
Sensor Model
Environment-/ Situation-analysis
Fusion
Localization
AD Core-Function
Planning
Driving Strategy
Control
longitudinal
lateral
lights
Open Simulation Interface (OSI) OpenScenario / OpenDRIVE
OSI
Obj-List
OSI Raw Data
Einviron-
ment Represen-tation
NDS-Format
Ground Truth
Traffic
Agents
Open-
SCENARIO
Open-
SCENARIO
Open-
SCENARIO
Open-
SCENARIO
Open-SCENARIO
actuator model
23
Simulation Architecture – Full Parameter Variation
OSI OSI
Obj-List
OSI Obj-List
OSI Raw Data
OSI Raw Data
OSI Data
OSI Data
OSI Data
2
Simulation Framework
Sto
cha
stic
Va
ria
tio
n
Open-
SCENARIO Open-SCENARIO
•Logical scenario •Parameter distributions •Metrics •Pass/fail criteria
Open-
DRIVE Open-
DRIVE
Open-
DRIVE
Te
st R
esu
lt E
valu
ati
on
24
PEGASUS Simulation Architecture
Layer 2: Traffic Infrastructure • Boundaries (structural) • Traffic signs, elevated barriers
Layer 1: Road-Level • Geometry, topology • Quality, boundaries (surface)
Layer 3: Temporary manipulation of Layer 1 and Layer 2 • Geometry, topology (overlaid) • Time frame > 1 day
Layer 4: Objects • Static, dynamic, movable • Interactions, maneuvers
Layer 5: Environment • Weather, lighting and other
surrounding conditions
Layer 6: Digital Information • (e.g. )V2X information,
digital map