Autonomous Vehicles (AVs): Basics and Testing Challenges

The session will be held as part of Exactpro’s ongoing AI Testing Talks event series. Session expert, Julia Emelianova, PhD, Researcher, Exactpro, will cover:
📌the motivation behind and the process of autonomous vehicle development,
📌the main principles of automated driving and the existing navigation challenges,
📌testing approaches and the main testing objectives for autonomous vehicles.

Follow us on
LinkedIn https://www.linkedin.com/company/exactpro-systems-llc
Twitter https://twitter.com/exactpro

Transcript

1. 1 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks BUILD SOFTWARE TO TEST SOFTWARE exactpro.com Autonomous Vehicles (AVs): Basics and Testing Challenges Julia Emelianova PhD, Researcher, Exactpro 14 JULY | 1:30 PM SLST
2. 2 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks Contents 1. Motivation for Building Autonomous Vehicles (AVs) 4 2. Autonomous Vehicles: History 5 2.1. Phase 1. Foundational Research 5 2.2. Phase 2. Grand Challenges 7 2.3. Phase 3. Commercial Development 8 3. Levels of Automation 9 4. Automated and Connected Driving 11 4.1. Internet of Vehicles (IoV) - Big Data Architecture 11 4.2. Vehicle-to-everything (V2X) Communication 12 4.3. Common AV Sensor Setup 13 4.4. Main Parts of the AV Navigation Process 16
3. 3 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks Contents 5. AV Navigation Challenges 17 5.1. Perception Challenges 17 5.2. Localization Challenges 18 5.3. Prediction and Decision Making Challenges 19 5.4. Data Processing Challenges 20 6. AV Testing 21 6.1. Test Instances Based on X-In-the-Loop (XiL) Approaches 21 6.2. AV Evolutionary Design and Testing Flow as a the V-Model 23 6.3. Main Testing Objectives 24 6.4. Examples of AV Simulators 25 6.5. Scenario-Based SiL Testing Approach 27 6.6. Examples for Attack Simulation 34 6.7. Metrics for AV Testing 38
4. 4 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 1. Motivation for Building Autonomous Vehicles (AVs) greater personal independence saving money increased productivity reduced traffic congestion environmental gains greater road safety Autonomous Vehicle (AV) is a self-driving car which moves safely with little or no human input.
5. 5 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 2. Autonomous Vehicles: History 1. The development of automated highway systems, in which vehicles depend significantly on the highway infrastructure to guide them: - follow magnets imbedded in the roadway to keep the line centering - track a radar reflective stripe, etc. https://www.ri.cmu.edu/pub_files/pub2/thorpe_charles_1997_2/thorpe_charles_1997_2.pdf https://www.youtube.com/watch?v=RlZEeIC_2lI 1980 - 2003, university research centers, sometimes in partnership with transportation agencies and automotive companies, undertook basic studies of autonomous transportation. Two main technology concepts emerged from this work: https://www.jstor.org/stable/10.7249/j.ctt5hhwgz.11?seq=2 National Automated Highway Systems Consortium Demo ‘97 in San Diego 2.1. Phase 1. Foundational Research
6. 6 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 2. Autonomous Vehicles: History 2. The development of both semi-autonomous and autonomous vehicles that depend little on the highway infrastructure: - vision-guided vehicles VaMoRs (1986), VaMP and VITA-2 (the Prometheus project, 1987-1995) constructed by team led by Ernst Dickmanns https://www.lifehacker.com.au/2018/12/today-i-discovered-the-self-d riving-car-trials-of-the-1980s/ - Navlab models 1-5 vehicles developed at Carnegie Mellon University (1984-1995) https://www.digitaltrends.com/cars/first-self-driving-car-ai-navlab-his tory/ VaMoRs, 1986 VaMP, 1995 Navlab models 1-5, 1984-1995 2.1. Phase 1. Foundational Research
7. 7 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 2. Autonomous Vehicles: History 2003 - 2007, DARPA Challenges: 1. 2004 - self-navigate 240 km of desert roadway (no car completed the route) 2. 2005 - self-navigate 212 km (5 cars completed the course) 3. 2007 - the urban challenge in a mock city environment (4 cars completed the route in the allotted 6-hour time-limit) During the development, better software, camera, radar and laser sensors improved the road following and collision avoidance. The autonomous system was sensing the environment and made the decisions. 2005 DARPA Grand Challenge https://en.wikipedia.org/wiki/DARPA_Grand_Challenge 2.2. Phase 2. Grand Challenges
8. 8 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 2. Autonomous Vehicles: History 2.3. Phase 3. Commercial Development Since 2007
9. 9 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 3. Levels of Automation https://www.sae.org/binaries/content/assets/cm/content/blog/sae-j3016-visual-chart_5.3.21.pdf https://en.wikipedia.org/wiki/Advanced_driver-assistance_system https://www.asam.net/index.php?eID=dumpFile&t=f&f=2776&token=6e8da9b58594ecefb20dcc06c0ac55480df14c67 SAE J3016 (published in 2014) The Society of Automotive Engineers (SAE) defines 6 levels of driving automation ranging from 0 (fully manual) to 5 (fully autonomous). Advanced Driver-Assistance System (ADAS) - any of the groups of electronic technologies that assist drivers in driving and parking functions.
10. 10 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 3. Levels of Automation Robotaxis with SAE Level 4 are now used for passenger transportation in several cities both in USA and China Mercedes-Benz becomes world’s first to get Level 3 autonomous driving system approved for European public roads starting from May 17 2022 https://www.therobotreport.com/mercedes-rolls-out-level-3-autonomous-driving-tech-in-germany/ https://waymo.com/waymo-driver/ https://www.autox.ai/en/index.html
11. 11 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 4. Automated and Connected Driving 4.1. Internet of Vehicles (IoV) - Big Data Architecture MAN - Metropolitan Area Network RFID - Radio Frequency IDentification https://www.sciencedirect.com/science/article/pii/S1877050918304083
12. 12 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 4. Automated and Connected Driving 4.2. Vehicle-to-everything (V2X) Communication https://www.thalesgroup.com/en/markets/digital-identity-and-security/iot/industries/automotive/use-cases/v2x
13. 13 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 4. Automated and Connected Driving 4.3. Common AV Sensor Setup - Inertial Measurement Units (IMU) or Inertial Navigation System (INS) - Global Positioning Systems (GPS) and Differential GPS (DGPS) - Odometry Sensors https://arxiv.org/pdf/1910.07738.pdf https://www.cpp.edu/~ftang/courses/CS521/notes/sensing.pdf https://www.ansys.com/content/dam/resource-center/article/ansys-advantage-autonomous-vehicles-aa-V12-i1.pdf (page 32)
14. 14 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 4. Automated and Connected Driving 4.3. Common AV Sensor Setup https://techgameworld.com/mercedes-drive-pilot-approved-in-germany/
15. 15 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks RADAR: - object detection at a long distance and through fog or clouds - resolution several meters at a distance of 100 m - noisy sensor LiDAR: - a high level of accuracy for 3D mapping - resolution of a few centimeters at a distance of 100 m Camera: - colour detection - unable to measure distances 4. Automated and Connected Driving 4.3. Common AV Sensor Setup https://www.fierceelectronics.com/components/lidar-vs-radar https://higherlogicdownload.s3.amazonaws.com/AUVSI/14c12c18-fde1-4c1d-8548-035ad166c766/UploadedImages/2017/PDFs/Proceedings/BOs/Bo6-1.pdf https://www.researchgate.net/figure/A-possible-combination-of-sensors-for-all-weather-navigation_tbl2_346038646/download
16. 16 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 4. Automated and Connected Driving 4.4. Main Parts of the AV Navigation Process https://neptune.ai/blog/self-driving-cars-with-convolutional-neural-networks-cnn https://www.sciencedirect.com/science/article/abs/pii/S0968090X18302134
17. 17 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 5. AV Navigation Challenges 5.1. Perception Challenges • Object labeling and classification • Segmentation • Correspondence problem • Scene understanding https://neptune.ai/blog/self-driving-cars-with-convolutional-neural-networks-cnn https://www.youtube.com/watch?v=aQwqD5cB2ck The same object Labeling tools, ML and AI approaches help to solve them
18. 18 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 5. AV Navigation Challenges 5.2. Localization Challenges • Positioning and Global localization challenges (GPS, INS) • Road lines, traffic signs detection https://www.jstage.jst.go.jp/article/essfr/9/2/9_131/_pdf/-char/en https://neptune.ai/blog/self-driving-cars-with-convolutional-neural-networks-cnn
19. 19 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 5.3. Prediction and Decision Making Challenges https://arxiv.org/abs/1912.11676 • Occupancy and flow prediction • Pedestrians behavior prediction • Self-driving car motion prediction Task examples: - adjusting speed when anticipating a curve - collision avoidance - follow a line or a path - speed control 5. AV Navigation Challenges
20. 20 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 5. AV Navigation Challenges 5.4. Data Processing Challenges https://intellias.com/the-emerging-future-of-autonomus-driving/ https://www.youtube.com/watch?v=lCohTPSFj3I (at 17:10) Large data volumes Ultra-high computing Identification problems Difficulties in finding a valuable data
21. 21 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.1. Test Instances Based on X-In-the-Loop (XiL) Approaches https://www.avl.com/documents/10138/2699442/GSVF18_Validation+of+X-in-the-Loop+Approaches.pdf/0b13c98a-7e6d-45e7-baab-2a8d80403c38 https://www.researchgate.net/publication/311919670_Autonomous_vehicles_testing_methods_review
22. 22 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.1. Test Instances based on X-In-the-Loop (XiL) Approaches https://www.avl.com/documents/10138/2699442/GSVF18_Validation+of+X-in-the-Loop+Approaches.pdf/0b13c98a-7e6d-45e7-baab-2a8d80403c38
23. 23 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.2. AV Evolutionary Design and Testing Flow as a V-Model https://www.researchgate.net/publication/311919670_Autonomous_vehicles_testing_methods_review
24. 24 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.3. Main Testing Objectives Description Purpose(s) Search for the best sensors setup (number of sensors and their positioning on the vehicle) Perception Check the correctness of sensor data annotation Scene understanding, labeling and segmentation Check the decision making, planning and control algorithm model Prediction and decision making Possible attack simulations and search for vulnerabilities Security research
25. 25 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.4. Examples of AV Simulators Open-source Simulator Owner SVL Simulator LG Electronics America R&D Centre Apollo Simulation Platform Baidu CARLA Simulator CARLA Team Udacity AV unity simulator Udacity SUMO Eclipse Foundation AirSim Microsoft SUMMIT PGDrive Closed-source Simulator Owner Prescan Siemens CarSim Mechanical Simulation Corporation Carcraft simulator Waymo Virtual Testing Suite platform Aurora VIRES VTD simulator VIRES Simulationstechnologie GmbH rFpro’s simulation software rFpro
26. 26 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks Scene understanding, labeling and segmentation testing 6. AV Testing 6.4. Examples of AV Simulators https://www.youtube.com/watch?v=NksyrGA8Cek VIRES VTD simulator CARLA simulator https://carla.readthedocs.io/en/latest/ref_sensors/ https://www.mdpi.com/2076-3417/12/1/281/htm
27. 27 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.5. Scenario-Based SiL Testing Approach Simulation Cycle https://www.claytex.com/tech-blog/automated-testing-methodologies-for-autonomous-vehicles/ Two main approaches for the scenario generation: 1. Manual 2. Automatic
28. 28 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.5. Scenario-Based SiL Testing Approach https://www.asam.net/standards/simulation-domain-overview/ In 2018 Association for Standardisation of Automation and Measuring Systems (ASAM) started the standardisation process to enable collaborative data development and data exchange between different tools and simulators and make autonomous vehicles testing more flexible and easy.
29. 29 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks ASAM recorded data and scenarios workflow 6. AV Testing 6.5. Scenario-Based SiL Testing Approach https://www.asam.net/project-detail/asam-openlabel-v100/
30. 30 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks Scenario generation is the process of taking a variable set from the Test Manager and creating a test case in simulation. Each simulation includes: • Ego Vehicle with the system under test, i.e. the controller • Road Network • Road Features • Traffic • Weather 6. AV Testing 6.5. Scenario-Based SiL Testing Approach Data Layer Model for Scenario Description https://www.pegasusprojekt.de/files/tmpl/Pegasus-Abschlussveranstaltung/15_Scenario-Database.pdf
31. 31 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.5. Scenario-Based SiL Testing Approach Example of Test Cases generation https://www.researchgate.net/publication/311919670_Autonomous_vehicles_testing_methods_review
32. 32 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.5. Scenario-Based SiL Testing Approach Example of Test Cases generation SVL Simulator
33. 33 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.5. Scenario-Based SiL Testing Approach Test areas: 1. Configuration of the system a. Environmental Conditions i. Weather Conditions ii. Road Conditions iii. Illumination b. Traffic Infrastructure i. Traffic speed / Driving Modes ii. Agents Diversity c. Physical Infrastructure i. Roadway Types ii. Roadway Surfaces iii. Roadway Geometry iv. Geographic Area d. Operational Constraints i. Speed Limit ii. Traffic Conditions 2. Configuration of the ego vehicle a. Vehicle model b. Sensor configuration 3. Traffic actions a. Safety scenarios (regular traffic actions) b. Traffic accidents and road emergencies
34. 34 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing https://dl.acm.org/doi/10.1145/3372297.3423359 Attack generation against sensors on the Perception and Prediction levels Attack against camera 6.6. Examples for Attack Simulation
35. 35 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.6. Examples for Attack Simulation https://dl.acm.org/doi/10.1145/3372297.3423359 Attacking Tesla via a digital billboard Attack generation against sensors on the Perception and Prediction levels Attack against camera
36. 36 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.6. Examples for Attack Simulation https://sites.google.com/view/cav-sec/planfuzz https://www.ndss-symposium.org/wp-content/uploads/2022-177-paper.pdf Semantic Denial-of-Service (DoS) vulnerabilities are most generally exposed in practical AD systems due to the tendency to avoid safety incidents. They arise from the program code for AD planning and decision making systems. Lane following DoS attack. In this scenario, the AD vehicle keeps cruising in the current lane while static or dynamic obstacles are located outside of the current lane boundaries. The victim AD vehicle permanently stops due to off-the-road cardboard boxes.
37. 37 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.6. Examples for Attack Simulation https://sites.google.com/view/cav-sec/planfuzz Lane Changing DoS Attack on Apollo. In this scenario, the victim AD vehicle gives up a necessary lane changing decision even though the lane it needs to change to is empty and the attacking vehicle following it shows no intention to change to that lane.
38. 38 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.7. Metrics for AV Testing https://www.researchgate.net/publication/348116055_Quality_Metrics_and_Oracles_for_Autonomous_Vehicles_Testing 1. Driving quality metrics based on mutual configuration parameters of: • ego vehicle • infrastructure elements (signs, borders, facilities) • other traffic participants (vehicles, bikes, pedestrians) Examples: speed, acceleration, position, steering, braking and collisions.
39. 39 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks 6. AV Testing 6.7. Metrics for AV Testing https://arxiv.org/abs/1912.11676 2. Evaluation Metrics for vehicle behavior prediction a. Intention prediction metrics: • Accuracy • Precision • Recall • F1 Score • Negative Log Likelihood (NLL) • Average Prediction Time b. Trajectory prediction metrics: • Final Displacement Error (FDE) • Mean Absolute Error (MAE) or Root Mean Square Error (RMSE) • Minimum of K Metric • Cross Entropy • Computation Time
40. 40 BUILD SOFTWARE TO TEST SOFTWARE AI Testing Talks AI Testing Talks Thank You! Questions?