Cost-effective Simulation-based Test Selection in Self-driving Cars SW
summary
A cost-saving pipeline called SDC-Scissor uses BeamNG simulation data and machine learning to predict which simulation scenarios are unlikely to reveal faults and skip them before execution. Tests are generated as road x y coordinate sequences, labeled by running them in BeamNG, and the dataset shows failing runs take about 137 seconds while most tests pass and run longer. A genetic algorithm then prioritizes selected tests to minimize execution time and maximize scenario diversity.
AI-written summary · about
chapters (26)
- 0:00Introduction
- 0:33About me
- 1:08Cyberphysical systems
- 2:00Mobility
- 2:58Reliability
- 4:36Use Cases
- 5:44Anomalies in Selfdriving Cars
- 6:30Tesla vs Private Jet
- 7:19Smart Summon
- 8:14Multimeter
- 10:37Testing
- 12:21Simulation
- 13:40Execution Time
- 15:02How do we select relevant test scenarios
- 15:30The SDCC tool
- 15:52Where to find it
- 16:26Demo
- 20:18Test Prioritization
- 21:58Collaboration
- 22:34Cosmos
- 23:42Questions
- 26:12What makes it safe
- 28:03The simulator
- 32:47Questions and answers
- 35:45Work in progress
- 39:20Test case
related talks
24mUse of AI to quantify the "safety of autonomous vehicles"
17mSimulation based method for the analysis of energy-efficient driving algorithms
30mHow to train your robot?
23mA TESTER’S LIFE IN THE AUTOMOTIVE INDUSTRY
31mDeep learning for self-driving vehicle technology
51mTeaching Autonomous Vehicles to Drive with Visible and Non-Visible Light Simulation
40mSmart cars of tomorrow: real-time driving patterns
38mFinding Lane Lines for Self Driving Cars