The total testbed area is approx 150 sq. meters and comprises of six separate locations. These include the main testbed with motion capture system and five locations that are in NLOS. The locations are general office areas and include like chairs, tables, glass door, electronics, metalshelves, etc. Additional description and details of the setup are available in our paper Toolbox Release: A WiFi-Based Relative Bearing Sensor for Robotics
This includes data samples across a total of ten positions of receiving (RX) robot arranged in a grid which are at a minimum distance of 2.5m from a Line-of-Sight (LOS) transmitting (TX) robot position. For Non-line-of-sight (NLOS), TX robots are placed at different positions in adjacent office spaces. A total of about 600 data samples were collected for this dataset.
The directory structure is as follows:
├── config_files
├── Dataset
│ ├── 2D-Displacement_output_files
│ │ ├── default_configuration
│ │ ├── low_and_sub_configuration
│ │ ├── low_res_configuration
│ │ └── sub_sample_configuration
│ └── 2D-Displacement_raw_data
│ ├── LOS_Set_A
│ │ ├── 0627_2D-multi-link_A_L1
│ │ ├── 0627_2D-multi-link_A_L10
│ │ ├── 0627_2D-multi-link_A_L2
│ │ ├── 0627_2D-multi-link_A_L3
│ │ ├── 0627_2D-multi-link_A_L4
│ │ ├── 0627_2D-multi-link_A_L5
│ │ ├── 0627_2D-multi-link_A_L6
│ │ ├── 0627_2D-multi-link_A_L7
│ │ ├── 0627_2D-multi-link_A_L8
│ │ └── 0627_2D-multi-link_A_L9
│ ├── LOS_Set_B
│ │ ├── 0628_2D-multi-link_B_L1
│ │ ├── 0628_2D-multi-link_B_L10
│ │ ├── 0628_2D-multi-link_B_L2
│ │ ├── 0628_2D-multi-link_B_L3
│ │ ├── 0628_2D-multi-link_B_L4
│ │ ├── 0628_2D-multi-link_B_L5
│ │ ├── 0628_2D-multi-link_B_L6
│ │ ├── 0628_2D-multi-link_B_L7
│ │ ├── 0628_2D-multi-link_B_L8
│ │ └── 0628_2D-multi-link_B_L9
│ ├── NLOS_Set_A
│ │ ├── 0627_2D-multi-link_NLOS_A_L1
│ │ ├── 0627_2D-multi-link_NLOS_A_L10
│ │ ├── 0627_2D-multi-link_NLOS_A_L2
│ │ ├── 0627_2D-multi-link_NLOS_A_L3
│ │ ├── 0627_2D-multi-link_NLOS_A_L4
│ │ ├── 0627_2D-multi-link_NLOS_A_L5
│ │ ├── 0627_2D-multi-link_NLOS_A_L6
│ │ ├── 0627_2D-multi-link_NLOS_A_L7
│ │ ├── 0627_2D-multi-link_NLOS_A_L8
│ │ └── 0627_2D-multi-link_NLOS_A_L9
│ ├── NLOS_Set_B
│ │ ├── 0629_2D-multi-link_NLOS_B_L1
│ │ ├── 0629_2D-multi-link_NLOS_B_L10
│ │ ├── 0629_2D-multi-link_NLOS_B_L2
│ │ ├── 0629_2D-multi-link_NLOS_B_L3
│ │ ├── 0629_2D-multi-link_NLOS_B_L4
│ │ ├── 0629_2D-multi-link_NLOS_B_L5
│ │ ├── 0629_2D-multi-link_NLOS_B_L6
│ │ ├── 0629_2D-multi-link_NLOS_B_L7
│ │ ├── 0629_2D-multi-link_NLOS_B_L8
│ │ └── 0629_2D-multi-link_NLOS_B_L9
│ └── NLOS_Set_C
│ ├── 2D-multilink-NLOS_L1
│ ├── 2D-multilink-NLOS_L10
│ ├── 2D-multilink-NLOS_L2
│ ├── 2D-multilink-NLOS_L3
│ ├── 2D-multilink-NLOS_L4
│ ├── 2D-multilink-NLOS_L5
│ ├── 2D-multilink-NLOS_L6
│ ├── 2D-multilink-NLOS_L7
│ ├── 2D-multilink-NLOS_L8
│ └── 2D-multilink-NLOS_L9
└── figs
- config_files : Contains the configuration parameters used to process the data
- 2D-Displacement_output_files : Processed output data in json format
- 2D-Displacement_raw_data: Raw data files collected for each location of receiving robot (location 1 to 10) for a given position configuration of 3 Transmitting robots (A,B or C).
Each location directory (e.g 0627_2D-multi-link_A_L10) contains the following raw file:
├── csi_rx_*.dat
├── csi_tx2_*.dat
├── csi_tx3_*.dat
├── csi_tx4_*.dat
├── groundtruth_positions.json
├── odom_rx_trajectory_*.csv
├── rx_trajectory_*.csv
├── t265_rx_trajectory_*.csv
- csi_rx* : The CSI data file collected by a receiving robot
- csi_tx* : The CSI data file simultanesouly collected on all transmitting robots
- rx_trajectory* : Groundtruth displacement of receiving robot obtained using optitrack motion capture system.
- t265_rx_trajectory* : Estimated displacement of the receiving robot obtained using VIO sensor (Intel t265 tracking camera).
- odom_rx_trajectory* : Estimated displacement using robot odometer.
The following plot shows aggregate results for the AOA estimation accuracy using estimated robot displacement in non-line-of-sight.
Additional results for AOA estimation performance are available at our Wiki Page
The transmitting robot positions are assumed to be know and we use the entire data for all the position configuration of the robots (i.e data for LOS and data for NLOS). The receiving robot can localize itself using the bearing angle calculated from our framework. We use the profile variance metric discussed in to reject outlying measurements.
Results Coming soon!!
Additional results for the dataset are available at our Wiki Page
If this dataset is is useful for your research publications, please cite our work.
- [1] Ninad Jadhav, Weiying Wang, Diana Zhang, Swarun Kumar and Stephanie Gil. Toolbox Release: A WiFi-Based Relative Bearing Sensor for Robotics.
@article{WSR_toolbox,
title={Toolbox Release: A WiFi-Based Relative Bearing Sensor for Robotics},
author={Ninad Jadhav and Weiying Wang and Diana Zhang and Swarun Kumar and Stephanie Gil},
journal={},
year={},
volume={}
}- [2] Ninad Jadhav*, Weiying Wang*, Diana Zhang, O. Khatib, Swarun Kumar and Stephanie Gil. WSR: A WiFi Sensor for Collaborative Robotics (* denotes co-primary authors)
@article{Jadhav2020WSRAW,
title={WSR: A WiFi Sensor for Collaborative Robotics},
author={Ninad Jadhav and Weiying Wang and Diana Zhang and O. Khatib and Swarun Kumar and Stephanie Gil},
journal={ArXiv},
year={2020},
volume={abs/2012.04174}
}