Thanks for joining the presentation. Please submit your answer to the question on Mentimeter while you are waiting.
How does the future of mobility look like?
Optimized Test Selection for Simulation-based Tests for Self-driving Cars Software
By Christian Birchler
Master Thesis Presentation
Advised by:
Prof. Dr. Davide Scaramuzza and Dr. Sebastiano Panichella
14. September 2022
Before we start with self-driving cars...
How autonomous are systems today?
Autolanding
Automatic Train Operation
"Autogardening"
Self-driving Cars
Are they reliable?
Testing is crucial for self-driving cars!
Real-world Testing
- Realism
- Reliability
- Cost
- Replicability
Simulation-based Testing
- Cost
- Replicability
- Realism
- Reliability
Is simulation-based testing THE solution?
It depends...
We have the following problem:
How to do efficient simulation-based testing without losing reliability?
Regression Testing
- Selection
- Prioritization
- Minimization
Self-driving cars Cost-effective test selector
Self-driving cars CoSt-effective test selector
Self-driving cars CoSt-effective test selector
Self-driving cars CoSt-effective test selector
Self-Driving Cars - CoSt-effective test selector
Self-Driving Cars - CoSt-effective test selector
How does a simulation-based test look like?
When is a test failing or passing?
Passed
Failed
SDC-Scissor uses machine learning.
We need data!
Dataset Generation
Dataset Generation
Dataset Generation
Dataset Generation
Dataset Generation
Dataset Generation
Dataset Summary
Test Outcome Prediction
Test Outcome Prediction
Test Outcome Prediction
Test Outcome Prediction
Test Outcome Prediction
Test Outcome Prediction
Test Outcome Prediction
Feature Extraction
SDC-Scissor Use Cases
Prediction Performance
Cost-effectiveness
SDC-Scissor APIs
Flaky Tests
4'756 tests executed 10 times
54 (1.14%) weak flaky tests
2 (0.04%) strong flaky tests
Conclusion
- Prediction of unsafe test scenarios
- Reduction of simulation time
- Challenge of flaky tests
Thanks!
Your feedback is important for me.
