A 2D laser rangefinder scans dataset of standard EUR pallets
In the past few years, the technology of automated guided vehicles (AGVs) has notably advanced. In particular, in the field of factory and warehouse automation, different approaches have been presented for detecting and localizing pallets inside warehouses and shop-floor environments based on the data acquired from 2D laser rangefinders. In mohamed2018detection (), we present a robust approach allowing AGVs to detect, localize, and track multiple pallets using machine learning techniques based on an on-board 2D laser rangefinder. In this paper, the data used in (mohamed2018detection, ; mohamed2017detection, ) for solving the problem of detection, localization and tracking of pallets is described. Furthermore, we present an open repository of dataset and code111https://github.com/EMAROLab/PDT to the community for further research activities. The dataset comprises a collection of 565 2D scans from real-world environments, which are divided into 340 samples where pallets are present, whereas 225 samples represent the case in which no pallets are present.
keywords:Pallets, 2D laser rangefinder, 2D range data, Automated guided vehicle
|Subject area||Machine Learning and Robotics Engineering|
|More specific subject area||Pallets identification based on 2D range data|
|Type of data||Text files of the raw range data provided by the sensor|
|2D bitmap-like images of the 2D range data|
|MAT-files containing the entire dataset|
|Implementation codes of our approach in mohamed2018detection ()|
|How data was acquired||2D laser rangefinder from SICK AG (Model: S3000 Pro CMS)|
|Data format||Files in text format *.txt|
|2D RGB images in *.png ( & pixels)|
|MAT-files in MATLAB format *.mat|
|Scripts in M-file and C-file formats *.m, *.cpp|
|Data source location||University of Genoa, Italy|
|Data accessibility||Dataset and codes are archived in a GitHub repository at:|
|Related research article||mohamed2018detection ()|
Value of the Data
The data allows researchers to understand and get an overview of our approach for solving the problem of detection, localization and tracking of pallets based on machine learning techniques using a 2D laser rangefinder (mohamed2018detection, ).
The raw range data provided by the rangefinder can be used by the research community for further research activities, which can provide a significant contribution to the literature in so far as benchmarking and code reuse are concerned.
In the past few years, the technology of automated guided vehicles (AGVs) has notably advanced, to the point that nowadays it even equips consumer-level products. In the field of factory and warehouse automation, different approaches have been presented for detecting and localizing pallets inside warehouses and shop-floor based on different technologies. In particular, laser rangefinders are a popular choice.
The data we are presenting have been captured using a 2D laser rangefinder. This section provides an introduction to the test environment where the data has been captured, as well as details on data collection.
1.1 Test Environment
The dataset was collected in one of the Robotics labs at the University of Genoa, Italy. In particular, the test area is a typical indoor environment with an area of , which includes pallets, multiple obstacles such as walls as well as other robots, and furniture, as shown in Figure 1.
As mentioned in mohamed2018detection (), the raw range data provided by the laser rangefinder at the time instant (i.e., each frame) represents the set of measured distances from the rangefinder to surrounding environment in the direction given by the angle .
where is the maximum number of range points acquired per frame, which is related to the angular sensor’s resolution. In our case, , as the maximum field of view and the resolution of the rangefinder are 190 and 0.25 , respectively.
Afterwards, the array of range data can be represented in polar coordinates (i.e., ) as well as Cartesian coordinates (i.e., and ), giving a binary image of the operating area’s floor plan, using (2) and (3).
In our experiments, the raw range data consists of 2D range scans, which can be divided into two main classes:
Class #1 represents the case of having a pallet located in the environment. The total number of scans acquired from the sensor, where the pallet is present, amounts to 340 samples. Each sample must be different from the others, considering the distance and angle of the rangefinder with respect to the pallet, and the fact that the environment is dynamic, i.e., subject to change.
Class #2 considers the case in which no pallets are present in the environment, and consists of 225 samples. Bearing this in mind, the case in which there is an obstacle in front of the pallet’s operating face (as described in Section 2), which may prevent the robot from successfully forking the pallet, is not considered as a sample of Class #1.
Figure 2 presents samples of the dataset provided by the laser rangefinder, after converting them to Cartesian coordinates.
In order to avoid a number of problems presented in realistic situations such as wrong data association, irregularities in the environment terrain, vehicle vibrations, and so on, the data is collected based on two sequential stages:
the first stage requires the user’s supervision, especially for the scenes belonging to Class #1. In particular, range data has been conveniently visualized using a standard package rviz222http://wiki.ros.org/rviz/Tutorials in the Robot Operating System software framework (i.e., ROS333http://www.ros.org/about-ros/) to ensure that the pallet is actually in the scene;
the second stage involves the acquisition and the saving of four successive frames instead of one frame for each sample and stored in a text file.
Table 1 shows an example of the raw range data captured using the rangefinder over four successive frames. Moreover, the entire dataset of the two classes can be found in the GitHub repository at two folders called Class1 and Class2. Afterwards, the average over the four frames is calculated, which gives again an array of scan points. Finally, the raw 2D data (expressed in polar coordinates) are converted to Cartesian coordinates which have been used by our method in (mohamed2018detection, ) to generate 2D bitmap-like images.
|Frame # ()|
|END of Frame #|
Frame # () 0 3.11 1 3.11 2 3.11 ⋮ ⋮ 100 2.26 101 2.28 ⋮ ⋮ ⋮ ⋮ 757 1.51 758 4.08 759 4.06 760 4.08 END of Frame # Frame # () 0 3.13 1 3.11 2 3.13 ⋮ ⋮ 100 2.26 101 2.28 ⋮ ⋮ ⋮ ⋮ 757 1.51 758 4.08 759 4.08 760 4.05 END of Frame # Frame # () 0 3.13 1 3.11 2 3.00 ⋮ ⋮ 100 2.23 101 2.28 ⋮ ⋮ ⋮ ⋮ 757 1.48 758 4.08 759 4.08 760 4.08 END of Frame # Index Range data 0 3.12 1 3.11 2 3.06 ⋮ ⋮ 100 2.252 101 2.28 ⋮ ⋮ ⋮ ⋮ 757 1.50 758 4.07 759 4.075 760 4.07
2 Experimental Setup and System Overview
This section provides detailed information about the experimental settings, including the necessary tools and the used software for performing the task of range data acquisition and objects detection.
There are two main components necessary for acquiring 2D range data in our case, which are a 2D laser rangefinder and a pallet. In our research, a standard EUR-pallet has been used, which is the standard European pallet format as specified by the European Pallet Association (EPAL)444https://en.wikipedia.org/wiki/EUR-pallet. The size of EUR-pallets is 1200 800 with a height of 144 . Moreover, during our experiments, we define as operating face of the pallet the one of narrow width. On that face there are two slots, each 227.5 wide, as shown in Figure 3.
In the context of mobile robotics applications, there are many ways to measure distances, one of the most used approaches is Time-of-Flight (TOF) systems which measure the delay until an emitted signal hits a surface and returns to the receiver. Thus, the true distance from the sensor to the surface can be estimated. This category of systems involves sensing devices such as a Laser Measurement System (LMS) or LIDAR, radars, TOF cameras, or sonar sensors, which send rays of light (e.g., laser) or sound (e.g., sonar) in the environment, which will then reflect and return to the sensor. Knowing the speed with which a ray propagates and using precise circuitry to measure the exact time when the ray was emitted and when the signal was returned, the distance can be estimated easily.
In our experiments, real-world range data for detecting, localizing, and tracking the pallets are acquired using a commercial 2D laser rangefinder from SICK, model S3000 Pro CMS555https://www.sick.com/ag/en/s3000-professional-cms-sensor-head-with-io-module/s30a-6011db/p/p31284, as shown in Figure 3. The sensor has a maximum range of 49 (20 at reflectivity), a resolution of 0.25 , a 16 refresh frequency, and an empirical error of 30 . The maximum field of view of the rangefinder is 190 , which is largely sufficient for the detection of objects in front of it. The sensor generates an array of distances, which can be expressed in polar and Cartesian coordinates as mentioned before, each array with a size of per scan. It is noteworthy that the 2D laser rangefinder must be mounted on the robot at the same level of the pallet, for avoiding false detections or mis-detections of the significant pallet features.
The laser rangefinder is connected to a PC through a RS422-USB converter, which has a transmission rate of 500 . The PC used for acquiring data from the sensor is equipped with an Intel® Core i5-4210U 1.70 CPU and 6 of RAM, and runs Ubuntu 16.04 64 bit. In particular, real-world data is acquired using an ad hoc ROS node developed in C++. For further detailed information about the ROS package that has been used for reading the scan data from the SICK S3000 Pro CMS, please refer to the s3000_laser@GitHub page.
The work by I. S. Mohamed was supported by a scholarship from the ERASMUS+ European Master on Advanced Robotics Plus (EMARO+) programme. The authors would like to thank M.Eng. Yusha Kareem for his helping in data collection process.
Conflict of interest
The authors declare that they have no conflict of interest relevant to this article.
Transparency document. Supplementary material
Transparency data associated with this article can be found in the online version at https://github.com/EmaroLab/PDT.
- (1) I. S. Mohamed, A. Capitanelli, F. Mastrogiovanni, S. Rovetta, R. Zaccaria, Detection, localisation and tracking of pallets using machine learning techniques and 2D range data, arXiv preprint arXiv:1803.11254 (2018).
- (2) I. S. Mohamed, Detection and tracking of pallets using a laser rangefinder and machine learning techniques, Master’s thesis, European Master on Advanced Robotics (EMARO), University of Genova (2017).