A collection layout maps and sensor maps of different environments.
This repository of maps has been collected for the verification of methods presented in the following publications.
- Saeed Gholami Shahbandi, Martin Magnusson, 2D Map Alignment With Region Decomposition, CoRR, abs/1709.00309, 2017. URL(code)
- Saeed Gholami Shahbandi, Martin Magnusson, Karl Iagnemma. Nonlinear Optimization of Multimodal Two-Dimensional Map Alignment With Application to Prior Knowledge Transfer, in IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 2040-2047, July 2018. doi: 10.1109/LRA.2018.2806439. URL(code)
This repository contains a collection of maps from four different environments. For each environment, a set of sensor-based occupancy-like map is provided. These sensor map are collected as 3D meshes with a Google Tango tablet with the Constructor application provided by Google.
Each 3D mesh is horizontally sliced and converted to occupancy-like map. It should be mentioned that due to instability of the Constructor application in handling big meshes, some maps (mostly office maps) only cover the upper half (along z-axis) of the environment in order to increase the coverage of individual maps. Sensor maps vary in their global consistency and coverage of the area. In addition, a single layout map (in bitmap format, occupancy-like) is also provided for each environment. This table provides details for each set, followed by a thumbnail overview of all maps.
| Name | Type | #sensor maps | #layout | location |
|---|---|---|---|---|
| E5 | office building | 14 | 1 | E building, Halmstad University, Sweden |
| F5 | office building | 14 | 1 | F building, Halmstad University, Sweden |
| HIH | apartment | 4 | 1 | Intelligent Home Environment, Halmstad, Sweden |
| KPT4A | apartment | 4 | 1 | a residential apartment in Halmstad, Sweden |
- E5:
- F5:
- HIH:
- KPT4:
Note on mesh to occupancy map conversion: Due to the absence of sensor's trajectory and pose, identifying the open space and the conversion from mesh format to occupancy map has been done manually in an interactive process. This process also included manual filtering (eg. noise removal). As a consequence, the conversion is not deterministically reproducible. Generating a pseudo trajectory-pose state for sensor over each map, could make it possible to setup an automated procedure for the conversion. Such a procedure also requires a 3D point cloud filtering to result in a smooth occupancy map.
In order to run accompanied scripts for ground truth annotations or visualization of the annotations, the following dependencies must be met:
numpy >= 1.10.2
matplotlib >= 2.0
PySide >= 1.2.1
scikit-image >= 0.12
opencv >= 2
Most dependencies (except for opencv) could be installed by:
git clone https://github.com/saeedghsh/Halmstad-Robot-Maps.git
cd Halmstad-Robot-Maps
pip install -r requirements.txt
Instructions on how to use scripts will come soon.
Distributed with a GNU GENERAL PUBLIC LICENSE; see LICENSE.
Copyright (C) Saeed Gholami Shahbandi