Kognitive Systeme

AmbiSense UHF RFID Benchmark Dataset for Location Fingerprinting and Robotic Self-Localization

Artur Koch, Philipp Vorst

On this page we provide the experimental data used in the paper submitted as

[1] Philipp Vorst, Artur Koch, and Andreas Zell. Efficient self-adjusting, similarity-based location fingerprinting with passive UHF RFID. In IEEE International Conference on RFID-Technology and Applications (RFID-TA2011), pages 160--167, Sitges, Barcelona, Spain, September 15-16 2011. IEEE. [ DOI | details | pdf ]

The datasets can serve as benchmarks and experimental data for location fingerprinting using passive UHF RFID. The setup is based on two types of RFID readers mounted on a mobile robot. Stationary passive RFID tags were placed along the sides of corridors through which the robot traveled while recording the RFID log files. First, the datasets contain measurements of two types of UHF RFID readers which comply with the standards ISO/IEC 18000-6C / EPC Class 1 Generation 2. Second, each measurement is annotated with its recording position, based on laser-based Monte Carlo localization with a mean accuracy of better than 10 cm. Odometry (measured displacement) as provided by the robot's base are additionally provided in the logs.

See also the terms of use below.

Dataset Download

Please note that by downloading any of the following files you accept the terms of use.

Dataset Description
SCITOS G5, Elatec SR-113, 30.0 dBm (1.6 MB) This dataset was recorded with an Elatec SR-113 RFID reader on board a SCITOS G5 robot. The transmission power was 30.0 dBm.
SCITOS G5, Impinj Speedway, 30.0 dBm (2.1 MB) This dataset was recorded with an Impinj Speedway RFID reader on board a SCITOS G5 robot. The transmission power was 30.0 dBm.
SCITOS G5, Impinj Speedway, 30.0 dBm (3.7 MB) This dataset was recorded with an Impinj Speedway RFID reader on board a SCITOS G5 robot. The transmission power was 22.5 dBm.

Data Format

 

Each dataset consists of 5 log files, each of which follows the principal CARMEN log file format. The provided data files contain four types of data record: parameters, odometry, RFID data, and ground truth positions. Each record represents one row of a log file and is formatted as shown in the table below. Columns of each row are separated by white spaces. Positions of the robot relate to the robot's point of mass.

Record type Column within row (record) Fixed value Description
Logged parameter  
  1 PARAM  
  2   Name of the parameter
  3   Value of the parameter
  4   Timestamp (in seconds since 1970)
  5 scitos Name of the robot
  6   Timestamp (in seconds since 1970)
RFID measurement  
  1 RFID  
  2 1.300000 or 1.400000 driver version, can be ignored
  3   reader type:
3=Elatec SR-113
4=Impinj Speedway)
  4-6 1 reserved
  7 0 reserved
  8   transmission power:
1.000000=30.0 dBm
0.500000=22.5 dBm
  9 5 reserved
  10 3 flags of antennas used (20+21 for left+right antenna)
  11 1 reserved
  12   number of detected tags (n)
  13+(i*6)   ID of detected tag i (i=0,...,n-1)
  13+(i*6+1)   Antenna with which tag i was detected (0=left antenna, 1=right antenna)
  13+(i*6+2)   Number of detections of tag i in current measurement (detection count)
  13+(i*6+3)   Peak RSS value of tag i (double-valued, in dBm; zero if not provided in case of Elatec reader)
  13+(i*6+4)   Timestamp of first detection of tag i in current measurement (double-valued, in seconds since 1970)
  13+(i*6+5)   Timestamp of last detection of tag i in current measurement (double-valued, in seconds since 1970)
  n*6+13   Timestamp of inquiry start time (double-valued, in seconds since 1970)
  n*6+14   Timestamp of inquiry end time (double-valued, in seconds since 1970)
  n*6+15   Global X coordinate of the robot (in meters), to be used as reference/actual recording position of the measurement
  n*6+16   Global Y coordinate of the robot (in meters), to be used as reference/actual recording position of the measurement
  n*6+17   Global heading (orientation on XY plane) of the robot (in radians), to be used as reference/actual recording position of the measurement
  n*6+18 to n*6+20 0.000000 empty
Odometry  
  1 ODOM  
  2   Odometric X coordinate of the robot (in meters, relative to some starting position)
  3   Odometric Y coordinate of the robot (in meters, relative to some starting position)
  4   Odometric heading (orientation on XY plane) of the robot (in radians, relative to some starting position)
  5   Forward velocity (in m/s)
  6   Rotational velocity (in radians/s)
  7   Acceleration
Reference positions  
  1 TRUEPOS  
  2   Odometric X coordinate of the robot (in meters, relative to some starting position)
  3   Odometric Y coordinate of the robot (in meters, relative to some starting position)
  4   Odometric heading (orientation on XY plane) of the robot (in radians, relative to some starting position)
  5   Global X coordinate of the robot (in meters, relative to the map)
  6   Global Y coordinate of the robot (in meters, relative to the map)
  7   Global heading (orientation on XY plane) of the robot (in radians, relative to the map)

The last three columns of RFID measurements, odometry, and reference positions contain:

  • Timestamp of measurement (double-valued, in seconds since 1970)
  • Type of robot: "scitos"
  • Timestamp of measurement (double-valued, in seconds since beginning of log file)

Note that the reference position records are actually redundant, since each RFID measurement is already stamped with a reference position.

Images

The following photographs illustrate how RFID tags were placed during the experiments:

Videos

The following videos illustrate how RFID data were recorded:

Video 1 (17 MB) This video shows how the robot is passing RFID tags.
Video 2 (15 MB) This video shows how the robot is remotely controlled in the corridor part of the environment.
Video 3 (40 MB) In this video the camera assumes the perspective of the robot and shows how it passes the RFID tags. The height of the camera is slightly below the RFID antennas.

Map of the Environment

The image below represents the original occupancy grid map relative to which the robot localized itself during the experiments. You can download it in CARMEN map format.

Motion Model Parameters

In the experiments in the abovementioned paper we applied the motion model proposed by Eliazar and Parr (Learning Probabilistic Motion Models for Mobile Robots, Austin I. Eliazar and Ronald Parr, Proceedings of the 21st International Conference on Machine Learning, ICML 2004). The following table lists the parameters which quantify the uncertainty of movements as required for predicting particle positions.

μCd μDd μTd μCt μDt μTt
-0.0144 1.0040 0.0052 0.0912 -0.0159 0.9959
σCd σDd σTd σCt σDt σTt
0.0720 0.1129 0.0608 0.0744 0.1129 0.2000

Links to Related Sites

  • Radish: The Robotics Data Set Repository (repository of various benchmark robotics datasets)

Terms of Use

We provide the datasets, information about the datasets, and the associated material (altogether subsequently denoted by the "Software") because we hope that they are useful to you. We have collected all data thoroughly and described them to the best of our knowledge.

Copyright (c) 2011 Chair of Cognitive Systems

Permission is hereby granted, free of charge, to any person downloading the Software, to use the datasets for research and academic purposes, including the right to publish experimental results obtained by using the data. The Software, however, must not be sold or redistributed.

Please cite the abovementioned paper in your own article(s) if you use the data.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.