Research map / 3D vision / 3D generation

Autonomous Driving Research Requires a Community-Driven Data Paradigm

NeurIPS 2026 · 3D vision › 3D generation · approach: 3D perception for autonomous driving

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Position paper arguing autonomous driving research needs a collaborative, community-driven data paradigm, supported by a literature survey of 647 real-world driving dataset papers and small preliminary evaluations of underused datasets.

Problem

Autonomous driving research concentrates on a few benchmark datasets (KITTI, nuScenes, Waymo) despite hundreds of community-released datasets, because of fragmentation, poor visibility, incompatible protocols, maintenance decay, and benchmark incentives.

Approach

Proposes treating driving data as shared community infrastructure via a platform with structured metadata, decoupled data/annotation contributions, community-sustained maintenance, and unified cross-dataset standards, with a feedback loop from model failures on underused data to future collection.

Key results

• Literature review identifies 647 real-world driving dataset papers (2012-2026) spanning nearly 50 countries, with citations following a long-tailed distribution.
• Qwen3-VL-8B-Instruct accuracy on nighttime light-fog frames in Seeing Through Fog is 9%, versus 56% during the day.
• Qwen3-VL-8B, Cosmos-Reason2-8B and Qwen3-VL-2B all score 4-9% on nighttime light fog, showing shared failure.

Surprising findings

• Qwen3-VL-8B predicts snow for 53 of 200 clear-weather frames, with 47 of those also predicting snow-covered road surface, suggesting label-convention mismatch.
• Nighttime light-fog accuracy is very low across all evaluated VLMs, even though dense fog at night is recognized at 46-65%.

Limitations

• Preliminary, qualitative evaluations on a few datasets and models; not systematic benchmarks.
• Citation counts used as a proxy for dataset usage, which is an imperfect measure acknowledged by the authors.
• Proposed platform is an early sketch without implementation or evaluation of its effectiveness.

Experimental setup

Datasets: KITTI Cityscapes nuScenes Waymo Open Dataset BDD100K Seeing Through Fog METEOR

Benchmarks: Seeing Through Fog weather tagging (800 frames)

Metrics: accuracy (%)

Builds on: KITTI nuScenes Waymo Open Dataset MMDetection3D OpenPCDet 123D

Closest papers by meaning

• The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset (NeurIPS 2026 · similarity 0.88)
• Bench2Drive: Towards Multi-Ability Benchmarking of Closed-Loop End-To-End Autonomous Driving (NeurIPS 2024 · similarity 0.88)
• Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.86)
• End-to-End Autonomous Driving Through V2X Cooperation (AAAI 2025 · similarity 0.86)
• Robust Autonomy Emerges from Self-Play (ICML 2025 · similarity 0.86)
• Multiagent Multitraversal Multimodal Self-Driving: Open MARS Dataset (CVPR 2024 · similarity 0.86)