LLVM-AD


About the Workshop

The 4th LLVM-AD Workshop at IEEE ITSC 2025 aims to keep IEEE ITS Society (ITSS) at the forefront of autonomous driving research by integrating advancements in Large Language Models (LLMs) and Visual Language Models (VLMs). The workshop targets researchers, industry professionals, and students from the ITSS community, as well as participants from other communities (e.g., CV, AI, NLP), making ITSS conferences the go-to place for them to present cutting-edge research accomplishments. The content of this workshop will focus on VLMs and LLMs for solving autonomous vehicle-related tasks such as prediction or planning, as well as Intelligent Transportation Systems related tasks like traffic signal control. It aligns with the main conference theme by addressing emerging ITS technologies and applications. The workshop aims to provide a platform for researchers to present novel ideas, discuss potential issues, and explore new research directions in the rapidly evolving fields of LLMs and VLMs.

Additionally, LLVM-AD will host a challenge based on the NuPlanQA dataset to assess the capabilities of language and computer vision models in addressing autonomous driving challenges.


Keynote Speakers

TBD


NuPlanQA Challenge

NuPlanQA is a Visual Question Answering (VQA) benchmark designed to evaluate multi-view driving scene understanding on the NuPlan dataset. The group who ranks the first will receive a 500$ award.

The benchmark covers three core skills:

  1. Road environment perception
  2. Spatial relation recognition
  3. Ego-centric reasoning

Each skill is divided into three subtasks, resulting in a total of nine subtasks.

We provide:

  • 1 million VQA pairs for training
  • 1,800 multiple-choice questions for evaluation

📂 The dataset and evaluation set are available here: NuPlanQA GitHub

Evaluation Metric: Accuracy

  • Participants must report accuracy for each of the three skills as well as the overall (total) accuracy.

Organizers