Intelligent Intersections
summary
Intersections concentrate crashes and fatalities and autonomous vehicles still underperform: reported testing shows about 5,000 miles per disengagement versus human ~200,000–500,000 miles per crash. Analysis of the Tempe Uber collision lists four concrete failures—phase prediction, unseen left-turn permission, blind spots and late operator reaction—and argues that phase-change countdown estimates, full-signal-phase broadcasts and blind-spot occupied message from intersection sensors would eliminate the spatial/temporal uncertainties that current on-vehicle vision and simulation miss.
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:57Why intersections?
- 2:13The promise of Automated Vehicles (AVS)
- 3:10AVs rely on deep learning and simulation
- 4:27AV safety record
- 6:03Uber AV Crash in Tempe, AZ on March 24, 2017
- 8:00Uber Crash Simulated
- 8:53Lessons from Uber crash
- 12:23Other intersection crash scenarios
- 13:05Limited line of sight of pedestrian
- 13:20Functions of intelligent intersection
- 14:12Signal Phase and Timing (SPT)
- 16:01Blind Zone Calculation: Conceptual Approach
- 17:05Right turn on red (RTOR) has 7 conflicting movements
- 18:04Conflict zones for right turn movement
- 18:17Resolve conflicts visually: Blind zones
- 19:54Resolve conflicts using SPaT message
- 21:12Uber crash case
- 23:16Avoiding Uber crash with 12V
- 23:26Red light violation
- 25:18How likely can a RTOR lead to accident with pedestrian?
- 27:48Uber fatal accident
related talks
21mAutonomous driving and handling the handover
24mUse of AI to quantify the "safety of autonomous vehicles"
41mMachine Ethics and Emerging Technologies
08mHuman–AI Collab for Reliable Automotive Control Systems
27mDriving Autonomous Vehicles Forward w/Real World Applications
42mAutomated Driving - Taking the Human Factors Researcher out of the Loop
38mDesigning Machine Learning for Humans
31mDeep learning for self-driving vehicle technology