Seeing Real-Time Traffic Problems
summary
Phones trade cheap and fast sensors for noisy, preprocessed positions: iOS often latches a single wrong coordinate after internal Kalman smoothing, Android reports larger uncertainty. Tap-in/tap-out logs let reconstruction find train locations, but official signaling logs sometimes omit late trains. Toll gantry data show speed collapses to ~20 mph while throughput barely falls; a 4% non‑RFID rate plausibly triggers backward shock waves. An incentive platform with 300k users yields longitudinal habit parameters (α, β) and embedded randomized spins that enable causal simulation and targeted nudges.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:30The Cathedral in the Bazaar
- 4:07The Experiment
- 6:28The Data
- 11:01The Cathedral
- 14:02Bakerloo Line
- 16:40Data Analysis
- 18:49Legacy Systems
- 20:12Illinois Tollway
- 20:59Traffic Graph
- 24:55Average Speed
- 27:30Billing Data
- 28:02Incentives
- 29:32Braindead model
related talks
44mOff to the Races with Android
25mErlang-based Desynchronised Urban Traffic Simulation (...)
38mA Busy Android App Developers' Guide to Perfetto
28mMobile Application Performance Done The Right Way
43mThe Invisible Door: Reliability Gaps in the Front End
30mHow to Create Sustainable Mobile Apps
21mMobile Application Performance Done The Right Way
17mSignal priority for improving fluidity and decreasing fuel consumption