Se ha denunciado esta presentación.
Utilizamos tu perfil de LinkedIn y tus datos de actividad para personalizar los anuncios y mostrarte publicidad más relevante. Puedes cambiar tus preferencias de publicidad en cualquier momento.

Safety and Validation of Autonomous Vehicles -- VECS 2019

693 visualizaciones

Publicado el

VECS 2019 keynote presentation
Hologram
UL 4600

Publicado en: Tecnología
  • Sé el primero en comentar

Safety and Validation of Autonomous Vehicles -- VECS 2019

  1. 1. Safety and Validation of Autonomous Vehicles VECS AV & Active Safety Gothenburg, Sweden April 3, 2019 © 2019 Edge Case Research Prof. Philip Koopman @PhilKoopman
  2. 2. 2© 2019 Edge Case Research  Perception safety approaches  Edge Cases as the limiting factor  The Heavy Tail Distribution Perception stress testing  Accelerate building the triggering event “zoo”  UL 4600: Autonomous Product Safety Standard  Transparent safety argument disclosure Overview [General Motors]
  3. 3. 3© 2019 Edge Case Research Validating an Autonomous Vehicle Pipeline Control Systems  Control Software Validation  Doer/Checker Architecture Autonomy Interface To Vehicle  Traditional Software Validation Perception presents a uniquely difficult assurance challenge Randomized & Heuristic Algorithms  Run-Time Safety Envelopes  Doer/Checker Architecture Machine Learning Based Approaches  ???
  4. 4. 4© 2019 Edge Case Research  Good for identifying “easy” cases  Expensive and potentially dangerous Brute Force AV Validation: Public Road Testing http://bit.ly/2toadfa
  5. 5. 5© 2019 Edge Case Research Validation Via Brute Force Road Testing?  If 100M miles/critical mishap…  Test 3x–10x longer than mishap rate  Need 1 Billion miles of testing  That’s ~25 round trips on every road in the world  With fewer than 10 critical mishaps …
  6. 6. 6© 2019 Edge Case Research  Safer, but expensive  Not scalable  Only tests things you have thought of! Closed Course Testing Volvo / Motor Trend
  7. 7. 7© 2019 Edge Case Research  Highly scalable; less expensive  Scalable; need to manage fidelity vs. cost  Only tests things you have thought of! Simulation http://bit.ly/2K5pQCN Udacity ANSYS
  8. 8. 8© 2019 Edge Case Research You should expect the extreme, weird, unusual  Unusual road obstacles  Extreme weather  Strange behaviors Edge Case are surprises  You won’t see these in testing  Edge cases are the stuff you didn’t think of! What About Edge Cases? https://www.clarifai.com/demo http://bit.ly/2In4rzj
  9. 9. 9© 2019 Edge Case Research  Novel objects (missing from zoo) are triggering events Need An Edge Case “Zoo” http://bit.ly/2top1KD http://bit.ly/2tvCCPK https://goo.gl/J3SSyu
  10. 10. 10© 2019 Edge Case Research  Where will you be after 1 Billion miles of fly-fix-fly?  Assume 1 Million miles between unsafe “surprises”  Example #1: 100 “surprises” @ 100M miles / surprise – All surprises seen about 10 times during testing – With luck, all bugs are fixed  Example #2: 100,000 “surprises” @ 100B miles / surprise – Only 1% of surprises seen during 1B mile testing – Bug fixes give no real improvement (1.01M miles / surprise) Why Edge Cases Matter https://goo.gl/3dzguf
  11. 11. 11© 2019 Edge Case Research Real World: Heavy Tail Distribution(?) Common Things Seen In Testing Edge Cases Not Seen In Testing (Heavy Tail Distribution)
  12. 12. 12© 2019 Edge Case Research  Need to find “Triggering Events” to inject into sims/testing The Heavy Tail Testing Ceiling
  13. 13. 13© 2019 Edge Case Research  Sensor data corruption experiments ML Is Brittle To Environment Changes Synthetic Equipment Faults Gaussian blur Exploring the response of a DNN to environmental perturbations from “Robustness Testing for Perception Systems,” RIOT Project, NREC, DIST-A. Defocus & haze are a significant issue Gaussian Blur & Gaussian Noise cause similar failures
  14. 14. 14© 2019 Edge Case Research False negative when in front of dark vehicle False negative when person next to light pole Stress Testing Perception  Augmenting images with noise highlights perception issues  Identifies systemic weaknesses even in absence of noise False positive on lane marking False negative real bicyclist
  15. 15. 15© 2019 Edge Case Research  Brittle perception behavior indicates Edge Cases  Data augmentation reveals triggering events Automatically Detecting Edge Cases
  16. 16. 16© 2019 Edge Case Research  Mask-R CNN: examples of systemic problems Example Triggering Events via Hologram “Red objects” Notes: These are baseline, un-augmented images. (Your mileage may vary on your own trained neural network.) “Columns” “Camouflage” “Sun glare” “Bare legs” “Children” “Single Lane Control”
  17. 17. 17© 2019 Edge Case Research  Disengagements only catch near-collision surprises Going Beyond Disengagements
  18. 18. 18
  19. 19. 19© 2019 Edge Case Research  Centered on a Safety Case  Credit for existing safety standards  Mix & match cross-standards techniques  Discourages questionable practices  “Unknowns” are first class citizens  Balance between analysis & field experience  Field monitoring required to feed back to argumentation  Assessment findings & field data used to update standard  Plan: public standard by end of 2019  Standards Technical Panel forming Spring 2019 to review draft UL 4600 – An Autonomy Safety Standard https://youtu.be/e9mmAjUPWJM
  20. 20. 20© 2019 Edge Case Research  Safety case sufficiency  Hazards and risk mitigation  Development process  Machine learning and “AI”  Verification & Validation  Dependability & security  Data safety  Non-driver human interface  Metrics & assessment  Tool qualification  Maintenance  Other lifecycle concerns  Field data feedback  Dealing with unknowns UL 4600 Draft Topics Include: UL4600.com
  21. 21. 21© 2019 Edge Case Research  Unhelpful approaches:  “We’re really smart. Trust us.”  Buzzword roll calls  Silence  Show top pieces of a safety case  Why should the public believe?  Transparency is essential!  Include:  UL 4600-style safety argumentation  Credible independent assessment  Add: why your road testing is also safe Better AV Safety Disclosures
  22. 22. 22© 2019 Edge Case Research  More safety transparency  Independent safety assessments  Industry collaboration on safety  Minimum performance standards  Share data on scenarios and obstacles  Safety for on-road testing (driver & vehicle)  Autonomy software safety standards  Traditional software safety … PLUS …  Accelerated perception testing  UL 4600 Autonomous Vehicle Safety Standard Ways To Improve AV Safety http://bit.ly/2MTbT8F (sign modified) Mars Thanks!

×