MOTIF Lab · USF

Project · Active

OpenLKA: open dataset of lane keeping assist from market autonomous vehicles

· PI · 2024-05

What it is

OpenLKA is the first open benchmark dataset of lane-keeping-assist (LKA) behavior collected from a wide range of production autonomous vehicles in real driving. Released in January 2025, it pairs CAN-bus signals with synchronized forward-facing video and lane-perception output, captured across multiple market vehicles under varied roadway and weather conditions in the Tampa, Florida region.

Why it matters

Lane-keeping assist is one of the most widely deployed ADAS features, but its real-world performance is opaque: most public benchmarks are built on simulated environments or single-vehicle test fleets, and the behavior of production LKA systems differs widely between manufacturers. Without an open, multi-vehicle dataset, researchers cannot calibrate microsimulations of mixed-autonomy traffic, evaluate disengagement risk, or model the variance in LKA-equipped vehicle behavior that capacity and safety analyses now have to contend with.

OpenLKA addresses this gap by capturing how production vehicles actually behave — not how a single test platform behaves — with the supporting signals (CAN, video, perception) needed to reverse-engineer LKA decision logic.

Data collection

The dataset was built during the lab’s summer 2024 data-collection campaign, which used rental cars to systematically sample a wide range of market vehicles. The same campaign also produced two companion datasets that build on shared instrumentation:

  • A CAN-level vehicle control dynamics dataset
  • An EV regenerative braking dataset

Connections

This project is a flagship release for the lab’s Vehicle technologies thread. It also feeds back into the Traffic flow theory thread — production-LKA car-following data is precisely what calibrating mixed-autonomy fundamental diagrams requires.

Project team

  • Hao Zhou (PI)
  • Yuhang Wang
  • Abdulaziz Alhuraish
  • Shengming Yuan
  • Shuyi Wang

Outputs & releases

  • Paper: Wang, Alhuraish, Yuan, Wang, Zhou. OpenLKA: an open dataset of lane keeping assist from market autonomous vehicles. arXiv:2501.03287
  • Lab GitHub: github.com/MOTIF-Lab (TODO: link directly to OpenLKA repo when ready)