CANTraj: CAN Signals Recover 88% of Driving Routes Even Without GPS Data
MOTIF-Lab · hf · 2026-11-16
Teams from University of South Florida, Stanford, Purdue, and NTU release CANTraj and CANTraj-Bench, showing that open driving datasets and usage-based insurance programs leak location via vehicle CAN signals even after GPS coordinates are removed.
- Dataset: 3,940 drives from 454 drivers pairing decoded CAN events with time-aligned GNSS trajectories; the benchmark covers 3,675 drives, 1,380 hours, 85,919 km.
- Key finding: with a known start and heading, adding steering to speed raises the online map filter's five-minute route recovery from 56% to 87%, with median displacement error of 43 m vs 921 m for map-free dead reckoning.
- Offline beam search recovers 88% (with heading) / 81% (without) of test routes; an LLM harness selecting among candidate routes recovers 79% on test and 83% few-shot.
- Six re-implemented attacks are compared under explicit information-access protocols; data and code are open-sourced. A direct warning to automakers and insurers relying on "no GPS = private" assumptions.
More from Safety
- Reuters: Nvidia chips pile up in Hong Kong as China withholds import approval — _weiping · 2026-11-16
- Windows 11 build caught probing for Codex, Claude, Cursor agents via ETW telemetry — ssh4net · 2026-11-16
- Strong, reproducible multimodal safety result — but paper not ready: submit or not? — mariyaivasileva · 2026-11-16
- AI slop papers published under Hinton, LeCun and Fei-Fei Li's names — 机器之心 · 2026-11-16
- GPT-6.2 reportedly delayed weeks over safety and alignment issues; OpenAI called a "safety company" — 量子位 · 2026-11-16
- Can workplace Copilot monitor and report employee productivity? — MiracleWhipHero · 2026-11-16