Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Survey on Datasets for Perception in Unstructured Outdoor Environments

Peter Mortimer, Mirko Maehlisch

TL;DR

This survey maps the landscape of publicly available perception datasets for unstructured outdoor environments in field robotics, focusing on 2D/3D sensing, synthetic data, terrain, forestry, place recognition, and robot platforms. It analyzes dataset characteristics, annotation policies, and sensor layouts to understand biases and scalability, highlighting trends toward larger, more granular annotations and the scarcity of cross-task benchmarks. The authors discuss the need for standardized evaluation servers and ontologies to harmonize labels across datasets, along with the potential of synthetic data and foundation models to reduce labeling effort. The work serves as a practical guide for practitioners to select appropriate datasets and identifies opportunities to improve interoperability and reproducibility in field robotics research.

Abstract

Perception is an essential component of pipelines in field robotics. In this survey, we quantitatively compare publicly available datasets available in unstructured outdoor environments. We focus on datasets for common perception tasks in field robotics. Our survey categorizes and compares available research datasets. This survey also reports on relevant dataset characteristics to help practitioners determine which dataset fits best for their own application. We believe more consideration should be taken in choosing compatible annotation policies across the datasets in unstructured outdoor environments.

Survey on Datasets for Perception in Unstructured Outdoor Environments

TL;DR

This survey maps the landscape of publicly available perception datasets for unstructured outdoor environments in field robotics, focusing on 2D/3D sensing, synthetic data, terrain, forestry, place recognition, and robot platforms. It analyzes dataset characteristics, annotation policies, and sensor layouts to understand biases and scalability, highlighting trends toward larger, more granular annotations and the scarcity of cross-task benchmarks. The authors discuss the need for standardized evaluation servers and ontologies to harmonize labels across datasets, along with the potential of synthetic data and foundation models to reduce labeling effort. The work serves as a practical guide for practitioners to select appropriate datasets and identifies opportunities to improve interoperability and reproducibility in field robotics research.

Abstract

Perception is an essential component of pipelines in field robotics. In this survey, we quantitatively compare publicly available datasets available in unstructured outdoor environments. We focus on datasets for common perception tasks in field robotics. Our survey categorizes and compares available research datasets. This survey also reports on relevant dataset characteristics to help practitioners determine which dataset fits best for their own application. We believe more consideration should be taken in choosing compatible annotation policies across the datasets in unstructured outdoor environments.
Paper Structure (10 sections, 2 equations, 7 figures, 5 tables)

This paper contains 10 sections, 2 equations, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Datasets
  3. 2D Semantic Segmentation
  4. 3D LiDAR Semantic Segmentation
  5. Synthetic Datasets
  6. Terrain Datasets
  7. Forestry Datasets
  8. Place Recognition
  9. Robot Platforms
  10. Discussion and Conclusion

Figures (7)

  • Figure : An example image of each dataset presented in the same order as in Table \ref{['tab:2d_segmentation_datasets']}. Click on any image to open the download page for each of these datasets. Listed are OFFSEDneigel_offsed_2021, Freiburg Forestvalada_deepscene_2016, TAS500metzger_tas500_2020, YCORmaturana_ycor_2018, BotanicGardenliu_botanicgarden_2024, CaSSeDsharma_CaSSeD_2022, MAVSsharma_CaSSeD_2022, SynPhoRestnunes_synphorest_2022, CaTsharma_CaT_2022, RELLIS-3Djiang_rellis-3d_2021, RUGDwigness_rugd_2019, GOOSEmortimer_goose_2024.
  • Figure : Images with the LiDAR points projected into the scenes of SynPhoRestnunes_synphorest_2022, RELLIS-3Djiang_rellis-3d_2021 and GOOSEmortimer_goose_2024. The LiDAR in SynPhoRest is modeled after a LiDAR scanner with a non-repetitive scanning pattern, leading to a different pattern in comparison to the rotating LiDAR scanner observed in RELLIS-3D and GOOSE.
  • Figure : An example image of each dataset presented in the same order as in Table \ref{['tab:terrain_datasets']}. Click on any image to open the download page for each of these datasets. Listed are TrailNethoveidar_trailnet_2018, ORFDmin_orfd_2022, Rally Estoniatampuu_estoniadriving_2023, Valehosseinpoor_vale_2021. CaTsharma_CaT_2022, Verti-Wheelersdatar_vertiwheelers_2024.
  • Figure : An example image of each dataset presented in the same order as in Table \ref{['tab:forestry_datasets']}. Click on any image to open the download page for each of these datasets. Listed are BarkNet 1.0carpentier_barknet_2018, SynPhoRestnunes_synphorest_2022, CanaTree100grondin_CanaTree100_2022, Montmorencytremblay_montmorency_2020, TimberSegfortin_timberseg_2022, FinnWoodlandslagos_finn_2023, ForTrunkDetdasilva_fortrunkdet_2021, SynthTree43kgrondin_SynthTree43k_2022.
  • Figure : An example image of each dataset presented in the same order as in Table \ref{['tab:2d_segmentation_datasets']}. Click on any image to open the download page for each of these datasets. Listed are OFFSEDneigel_offsed_2021, Freiburg Forestvalada_deepscene_2016, TAS500metzger_tas500_2020, YCORmaturana_ycor_2018, BotanicGardenliu_botanicgarden_2024, CaSSeDsharma_CaSSeD_2022, MAVSsharma_CaSSeD_2022, SynPhoRestnunes_synphorest_2022, CaTsharma_CaT_2022, RELLIS-3Djiang_rellis-3d_2021, RUGDwigness_rugd_2019, GOOSEmortimer_goose_2024.
  • ...and 2 more figures