Research Analyzer
← Back ICRA 2026

Advancing Off-Road Autonomous Driving: The Large-Scale ORAD-3D Dataset and Comprehensive Benchmarks

Hanzhang Xue, Hao Fu, Yiming Nie, Qi Zhu, Liang Xiao, Dawei Zhao, Yu Hu

PDF

AI summary

Key figure (auto-extracted from paper)
ORAD-3D is the largest and most diverse off-road autonomous driving dataset, paired with a comprehensive benchmark suite to accelerate robust perception and planning research.
off-road autonomous driving large-scale dataset 3D occupancy prediction free-space detection vision-language models world model

Problem

Progress in off-road autonomous driving is hindered by a severe lack of large-scale, high-quality datasets and standardized benchmarks that capture the extreme variability of unstructured terrains, weather, and lighting conditions.

Approach

The authors collect and release ORAD-3D, a 57,000-frame multi-sensor dataset spanning diverse off-road environments and extreme conditions, alongside a unified benchmark suite evaluating five core perception and planning tasks.

Key results

  • Release of ORAD-3D, the largest off-road autonomous driving dataset to date
  • Comprehensive coverage of diverse terrains, extreme weather, and varying illumination
  • Establishment of standardized benchmarks for 2D free-space detection, 3D occupancy, path planning, VLM driving, and world modeling
  • Public release of dataset and code to enable reproducible research

Why it matters

Provides a critical foundational resource for researchers and engineers developing robust perception and navigation systems for unstructured, real-world off-road environments.

Abstract

A major bottleneck in off-road autonomous driv- ing research lies in the scarcity of large-scale, high-quality datasets and benchmarks. To bridge this gap, we present ORAD-3D, which, to the best of our knowledge, is the largest dataset specifically curated for off-road autonomous driving. ORAD-3D covers a wide spectrum of terrains—including wood- lands, farmlands, grasslands, riversides, gravel roads, cement roads, and rural areas—while capturing diverse environmental variations across weather conditions (sunny, rainy, foggy, and snowy) and illumination levels (bright daylight, daytime, twi- light, and nighttime). Building upon this dataset, we establish a comprehensive suite of benchmark evaluations spanning five fundamental tasks: 2D free-space detection, 3D occupancy prediction, rough GPS-guided path planning, vision–language model–driven autonomous driving, and world model for off- road environments. Together, the dataset and benchmarks provide a unified and robust resource for advancing perception and planning in challenging off-road scenarios. The dataset is publicly available at https://github.com/chaytonmin/ ORAD-3D-Dataset-For-Off-Road-AD.

Index terms

Data Sets for Robot Learning Vision-Based Navigation Field Robots

Related papers