Towards Long-term Robotics in the Wild

TL;DR

This paper introduces large-scale, multi-modal datasets for natural environment robotics, addressing a gap in existing datasets primarily focused on urban settings, and demonstrates their utility in various perception tasks.

Contribution

The paper presents WildPlaces and WildScenes, new benchmark datasets for natural environments, enabling improved research in long-term field robotics.

Findings

Datasets include synchronized image, lidar, semantic, and pose data.

Demonstrated utility in place recognition and semantic segmentation.

Addresses the scarcity of natural environment datasets.

Abstract

In this paper, we emphasise the critical importance of large-scale datasets for advancing field robotics capabilities, particularly in natural environments. While numerous datasets exist for urban and suburban settings, those tailored to natural environments are scarce. Our recent benchmarks WildPlaces and WildScenes address this gap by providing synchronised image, lidar, semantic and accurate 6-DoF pose information in forest-type environments. We highlight the multi-modal nature of this dataset and discuss and demonstrate its utility in various downstream tasks, such as place recognition and 2D and 3D semantic segmentation tasks.