RoboLoc: A Benchmark Dataset for Point Place Recognition and Localization in Indoor-Outdoor Integrated Environments

TL;DR

RoboLoc introduces a LiDAR-based benchmark for place recognition that spans indoor and outdoor environments with continuous transitions and multi-floor navigation. It provides a large set of submaps generated from a 3D LiDAR, ground-truth trajectories from GLIM in LiDAR-only mode, and a strict train/test protocol to evaluate cross-domain localization. The authors benchmark PointNetVLAD, MinkLoc3D, and PTC-Net, revealing strengths of voxel-based and global descriptors in different regions and the limitations of transformer-based approaches in mixed domains. The dataset addresses a critical gap in robust localization for service robots and AMRs operating in GPS-denied campuses and buildings, and it supports future extensions to semantics and lifelong learning. RoboLoc thus serves as a realistic, scalable testbed for geometry-centric localization under domain shifts.

Abstract

Robust place recognition is essential for reliable localization in robotics, particularly in complex environments with frequent indoor-outdoor transitions. However, existing LiDAR-based datasets often focus on outdoor scenarios and lack seamless domain shifts. In this paper, we propose RoboLoc, a benchmark dataset designed for GPS-free place recognition in indoor-outdoor environments with floor transitions. RoboLoc features real-world robot trajectories, diverse elevation profiles, and transitions between structured indoor and unstructured outdoor domains. We benchmark a variety of state-of-the-art models, point-based, voxel-based, and BEV-based architectures, highlighting their generalizability domain shifts. RoboLoc provides a realistic testbed for developing multi-domain localization systems in robotics and autonomous navigation

Figures (8)

• Figure 1: The mobile data collection platform used in the RoboLoc dataset consists of an AgileX Scout Mini robot base equipped with an OS1-32 LiDAR sensor. This platform was deployed to navigate various indoor and outdoor areas across the Ajou University campus, serving as the primary system for collecting spatial data used to generate the RoboLoc maps.
• Figure 2: Comparison of LiDAR maps recorded on a weekend (left) and a weekday (right) in the same parking area. Noticeable differences in the presence of parked vehicles highlight the temporal variation of static structures. These differences present challenges for consistent place recognition, emphasizing the need for temporal robustness.
• Figure 3: A visualization showing the actual environment mapped using LiDAR data, along with the corresponding robot trajectory, overlaid on a high-resolution satellite image of Ajou University. The LiDAR map captures the 3D structure of the campus, while the trajectory is illustrated as a simplified linear path to provide a clear overview of the robot’s movement throughout various indoor and outdoor areas.
• Figure 4: Overlay of two GLIM-generated trajectories recorded at different times. The dominant magenta color indicates strong spatial consistency, as the red and blue paths nearly coincide throughout the entire route.
• Figure 5: Representative LiDAR maps illustrating the environmental diversity of our data collection sites. The first row includes (from left to right): a forest trail, a densely structured outdoor facility, a plaza area, and a parking lot. The second row showcases a variety of road scenes, including intersections and building-surrounded streets. This diversity in natural, architectural, and road environments provides challenging and realistic settings for evaluating place recognition and localization algorithms.