Conference PaperPDF Available

Human-robot co-working system for warehouse automation

Human-robot co-working system for warehouse
Rafael Rey, Marco Corzetto, Jose Antonio Cobano, Luis Merino and Fernando Caballero,,,,
Service Robotics Laboratory
Universidad Pablo de Olavide
Seville, Spain
Abstract—This paper addresses the material handling problem
(MHS) in warehouse automation by proposing a system that uses
an automated guided vehicle (AGV) in industrial environments.
The aim is to optimize the picking task with respect to manual
operation in a paint factory. The work describes the whole system
to perform all the automatic tasks. The process is controlled by
the Manufacturing Process Management System (MPMS) and
an autonomous co-worker robot execute the mission in partially
known environments. The navigation system implemented is safe
and robust. It considers the people detection and unknown
static obstacles. Also, an ultra-wide-band localization system is
implemented by offering new capabilities for situation awareness
in factories. Experiments with a real holonomic platform called
ARCO are performed to validate the approach.
Index Terms—AGV, warehouse automation
In the last years, the industrial robotics in the smart factories
of the future is being fostered. Advances of smart sen-
sors, electric and electronic technologies, and manufacturing
technology are conducting to the intelligent manufacturing
technology. Currently, Europe, United States and China are
paying more attention to this area [1] [2] [3]. The International
Federation of Robotics (IFR) estimates that by 2019 more than
1.4 million new industrial robots are being installed in factories
around the world [4].
The smart factory, also known as Industry 4.0 [5], represents
a leap forward from more traditional automation to a fully
connected and flexible system. Concretely, Industry 4.0 refers
to a phase in the industrial production evolution that focuses
heavily on inter-connectivity, automation, machine learning,
and real-time data processing.
Therefore, the current context requires more and more intel-
ligent robotics solutions for the collaboration between humans
and robots. Particularly, mobile robots play an important role
to successfully enable key improvements in the manufacturing
Automated guided vehicles (AGVs) are an example of
efficient warehouse systems by replacing human with mobile
robots or collaborating with them. The popular Kiva robot [6]
from Amazon is a good example, although other system can
This work is partially supported by the ARCO experiment funded by the
HORSE project (Grant number 680734) under the Horizon 2020 Framework
be also found in the market. These systems are used in large-
sized factories.
Different approaches have been presented in the literature
for AGV on intelligent warehouses, many of them based on
forklifts. Simulated forklift AGV systems with focus on the
path planning problem is presented in [7]. It allowed that
robotic forklifts executed transportation tasks in an intelli-
gent warehouse-like environment. All was done in simulation
and authors did not consider unknown obstacles like people
walking or other vehicles moving in the warehouse. A large
number of studies has been done in [8] including AGV design,
flowpath layout design, collision and deadlock avoidance,
route selection and scheduling problem. The work presented
in [9] considered several robots and compare two collection
strategies in simulation. In [10], a navigation system of a
flexible AGV intended for operation in partially structured
warehouses is presented. However, person detection and track-
ing is not considered by the approach.
Generally, the factory space is split between human and
robot areas. This is counterproductive to perform coordinated
joint human-robot tasks. Moreover, the autonomous navigation
areas or paths of the AGVs are constrained to spaces with
reduced presence of people, reducing the efficient of the
This paper addresses the material handling problem (MHS).
AGVs are widely chosen by companies to implement robust,
safe and truly MHS. The work presents a human-robot co-
working systems for Small and Medium-sized enterprises
(SMEs) enterprises in partially known and unstructured envi-
ronments. The system will be validated in the non-automated
production line of a factory which produces water-based
paints. The main objective of this system is to optimize the
picking task with respect to manual operation. Nowadays, this
operation is carried out by an operator with a cart going to
the different warehouse sections to pick-up the material in the
right quantities for later insertion in the mixing tanks. One of
the advantages of the system proposed will be to improve the
carry of the materials with the AGV and avoid risky situations
of the worker by reducing the worker exposure during the
The main contribution of this paper is the development of a
system able to perform all the automatic tasks without requir-
ing major changes in manufacturing process. The warehouse
Fig. 1: ARCO robot used in the experiments (left) and current
ARCO robot to use in a factory (right).
process is controlled by a Manufacturing Process Management
System (MPMS) and it has been designed using the Business
Process Model and Notation (BPMN v2.0) which, as the
name suggests, it is a standard graphic notation for business
processes modelling, maintained by the Object Management
Group1. BPMN is generally used to model processes for
administrative, financial or insurance businesses. The use of
BPMN to model industry processes has been proposed by [11].
An implementation of BPMN on a smart industry 4.0 factory
has been described in [12], while in [13] a BPMN solution
has been applied for taking into account the human physical
risks in manufacturing processes.
The autonomous co-worker robot used is known as ARCO
robot (see Fig. 1). The robot will operate in conjunction with
the workers to collect the raw materials associated with the
production process, on the transportation of these materials
and assisting on the scheduling of mixing quantities and
timings. Also, an ultra-wide-band (UWB) localization system
for people detection has been implemented. The inclusion of
this localization system offers new capabilities for situation
awareness in factories.
The main contributions of the paper are summarised bellow:
1) Implementation of an autonomous AGV navigation sys-
tem able to operate in a factory without the need of
cables, line-painting or RFID installation in the floor.
2) Implementation of ultra-wide-band localization system
for people detection that will provide new capabilities
for robot situation awareness.
3) Low computation time of the algorithms implemented
in order to ensure safety when obstacles or unexpected
events take place.
4) Experimental validation with a real platform.
The paper is organised into seven sections. Section II
presents the software architecture of the system proposed. The
navigation of the AGV is described in Section III. Section IV
shows the UWB localization system and Section V the MPMS
Fig. 2: ARCO ROS software architecture.
application. Experiments and results are presented in Section
VI. Finally, conclusions are exposed in Section VII.
The aim of the proposed architecture is to provide the
required functionalities for AGV autonomous navigation in
a factory, adapting to possible changes of the environment.
Figure 2 shows the ROS architecture of the AGV navigation
module and People detection and tracking module of the
proposed system. It allows the AGV (henceforth ARCO) to
execute transportation tasks in an intelligent warehouse-like
To the software architecture has been designed following
the guideline given by the HORSE Architecture Team (see
[14] for more details) . The process is controlled by a Java
application, the Manufacturing Process Management System
(MPMS), which is running on a laptop acting as a server.
The application has been developed over a Java Spring Boot
framework 2, because it allows dependency injection, and it
has an embedded Tomcat servlet. The MPMS has an embedded
Camunda 3, an open source platform which gives to the BPMN
model the ability of being executed and to control in real time
the whole manufacturing process. Every task defined on the
designed model can execute scripts or Java code. Furthermore,
Camunda offers an embedded web application that allows
monitoring and managing the process execution in real time,
to watch variables values and to interact with the application
through forms.
The manufacturing orders are received by email. They
contain the list of raw materials, together with the amount that
are needed for each of them. Then, the navigation goals are
given one by one by the MPMS application to ARCO. The
robot does not receive information about the manufacturing
order or the sequence of goals it has to reach, all the process
handling is performed by the MPMS application. Once a
goal is received by ARCO, it should navigate efficiently and
Fig. 3: ARCO communication architecture.
safely in the environment. During the process, the MPMS
continuously checks the robot status.
The workers will interact with the system through a web
application they can access in their phones or tablet. They can
use forms to verify the amount of raw materials needed and
they can interact with the system sending confirmations when
their tasks are accomplished.
The communication between the MPMS and the ARCO
robot has been set up through a web service server developed
in Java which runs on the ARCO server (see Fig. 3). The robot
uses two ROS modules for communication: the ROS bridge
node acts as a web service client, and the ROS communication
node acts as an interpreter that translates the web server
messages into ROS topics, and vice versa.
This section presents the software developed to perform
robust autonomous navigation in the environment considered.
The basic ARCO’s autonomy consists of localization, per-
ception, path planning and path tracking. These modules are
detailed in next paragraphs.
A. Robot Localization Module
In order to perform autonomous navigation, a map of the
environment and a method to localize the robot in it are
needed. Localization plays an important role in mobile robots.
This module localizes the robot in a previously generated map
of the environment and provides this information to the Global
and Local Planner modules (see Fig. 2).
The sensors used are one laser placed in the font and another
in the rear part of the robot. A static map of the environment
should be computed during robot deployment. This laser-
based map is generated by using the gmapping ROS package
[15]. This package provides laser-based SLAM (Simultaneous
Localization and Mapping), so a 2-D map is built from both
lasers and pose data collected by ARCO.
The robot localization is based on the AMCL (Adaptive
Monte Carlo Localization) which is a probabilistic localiza-
tion system for a robot moving in 2D [16]. Given a map
of the environment, this module estimates the position and
orientation of a robot as it moves and senses the environment.
The algorithm makes use of a particle filter to represent the
distribution of likely states, with each particle representing a
possible state, i.e., an hypothesis of where the robot is. The
odometry estimates the pose of the robot by integrating the
robots internal states such as speed and steering angle, which
are computed from the wheel’s optical encoders readings. The
AMCL then compensates the error of this estimation, which
is called odometry drift.
B. Global Planner
Once a reliable localization of the robot and a goal is
provided, a trajectory is generated by the Global Planner.
This global trajectory is defined by a sequence of points.
Note that the Global Planner module will interact with the
MPMS, moving the robot to the different places where to pick
the materials. Therefore, inputs of the Global planner are the
localization of the robot, goal to reach and costmap of the
static map (global costmap).
The developed Global Planner is based on a variation of
the any-angle path planning algorithm Theta* called Lazy
Theta* [17]. Lazy Theta* is an optimization of Theta* which
reduces the number of line of sight checks. An advantage
of this algorithm is the optimization of the computational
load. Other advantages of this graph-based algorithm with
respect to sampling-based algorithms is that its result is closer
to the shortest path and its solution is always the same.
Furthermore, it gets a very low computation time and a high
repeatability. The last advantages make this algorithm suitable
for the proposed system by ensuring safety and robustness.
Algorithm 1 shows the pseudo-code of the Lazy Theta*
algorithm. The algorithm is based on the A* algorithm, and
this last one is basically the Dijkstra algorithm with Euclidean
distance as heuristic. Theta* takes into account the line of sight
between neighbours allowing any-angle paths, and finally Lazy
Theta* reduces the line of sight checks improving computation
times by keeping the same paths as Theta*.
Lazy Theta* starts expanding the first vertex and calculates
the gand hvalues of its visible neighbours. After this step,
it repeats the same procedure for every unexpanded visible
neighbour sand checks if these neighbours have line of sight
with the parent of his parent (start vertex at this moment).
Instead of checking the line of sight between two vertexes s
and s0, Lazy Theta* assumes that they do have line of sight
and at first it considers only Path 2 (ComputeCost in line 29),
it checks later if there exists line of sight in the SetVertex
procedure (line 34) and only changes the g-value and parent
2open := closed := ;
3g(sstart) := 0;
4parent(sstart):=sstar t;
5open.Insert(sstart,g(sstar t) + h(sstart ));
6while open 6=do
7s := open.Pop();
9If s=sgoal then
10 return ”path found”;
11 closed := closed s;
12 foreach s0nghbrvis(s)do
13 If s06∈ closed then
14 If s06∈ open then
15 g(s0) := ;
16 parent(s0) := NULL;
17 UpdateVertex(s,s0);
18 return ”no path found”;
19 end
20 UpdateVertex(s,s’)
21 gold := g(s0);
22 ComputeCost(s,s’);
23 If g(s0)< gold then
24 If s0open then
25 open.Remove(s’);
26 open.Insert(s0,g(s0) + h(s0));
27 end
28 ComputeCost(s,s’)
29 /*Path 2 */
30 If g(parent(s)) + c(parent(s), s0)< g(s0)then
31 parent(s0) := parent(s);
32 g(s0):= g(parent(s)) + c(parent(s), s0);
33 end
34 SetVertex(s)
35 If NOT lineofsight(parent(s),s) then
36 /*Path 1 */
37 parent(s) := argmin
(s0nghbrvis closed(g(s0) + c(s, s0)));
38 g(s) := min
(s0nghbrvis closed(g(s0) + c(s, s0)));
39 end
Algorithm 1: Lazy Theta*
of sif there is no line of sight. The same procedure is repeated
until the checked vertex is the goal vertex. For more detailed
information see [17].
Figure 4 shows some of the advantages of Lazy Theta*
algorithm in a complex scenario where the solution is difficult
to find. In this case the maze shown has a lot of dead ends.
For this non-realistic extreme case Lazy Theta* computes the
optimal shortest path in 3 seconds.
C. Local Planner
The Local planner makes use of a segment of the global
trajectory provided by the Global planner. This planner plays
Fig. 4: Example of trajectory computed by the Lazy Theta*
algorithm. The initial position (black square) is in the top and
the final position at the bottom of the image.
an important role in mobile robotics to re-plan quickly a
safe trajectory when an unexpected event takes places such
as unknown obstacles, persons appearing nearby the robot,
etc. Moreover, this module should provide a free-collision
trajectory to the Path tracker module.
A smarter AGV navigation should take into account the
detection of unknown obstacles into the trajectory in advance
for re-planning in real time thanks to a precise obstacle
positioning knowledge. The detection is achieved from the
measurements of the frontal and rear lasers, and the generation
of the local costmap. Obstacle avoidance and the following of
the global trajectory is performed by the Local Planner.
Therefore, inputs of the Local Planner are: a part of the
global trajectory to follow, the costmap of a local map around
the AGV generated from both lasers (local costmap), and
localization of the robot. The output is the trajectory, defined
as a sequence of waypoints, that the robot should follow. Note
that the local costmap is centered in the ARCO robot and its
dimension can be set up. In order to ensure safety and a quick
reaction of the robot, the frequency of the Local Planner is
high (about 20Hz).
The implemented Local planner is well adapted to changes
of the environment thanks to the local costmap generated from
the lasers and the efficient and fast path planning algorithm.
This algorithm is also based on Lazy Theta*. It considers a
reduced costmap and only a part of the global trajectory gen-
erated by the Global Planner. Figure 5 shows both global and
local trajectory. Local trajectory is computed by considering
the local costmap (square centered in the robot position). Note
that the obstacles detected in the local costmap are inflated (in
light blue) in order to consider a safety margin.
Fig. 5: Global trajectory (green square) and local trajectory
(red square) computed by the Local Planner. Local costmap
around the robot position is also shown.
Fig. 6: Navigation modes implemented.
D. Path tracker
The Path tracker module carries out the navigation of the
robot. It computes the linear and angular velocity commands
to navigate and follow the trajectory computed by the Local
planner. For the sake of safety, an emergency stop system
has also been implemented in the Path tracker module. It is
continuously checking if some obstacles (for example people
crossing) whose distance is lower than a threshold is detected
by the lasers. When this happens, ARCO immediately stops
and it moves again only when the detected obstacle has moved
away a distance greater than a second threshold.
There are two navigation modes implemented. The first one
sends commands to move ARCO in straight line, straight line
and slow rotation, and rotation in place, depending where the
next point is. The second one considers the omnidirectional
capabilities of ARCO. Both modes take into account the inertia
of the robot to perform safe movements.
Figure 6 shows a test by using each navigation mode. The
position provided by AMCL and the heading of ARCO robot
are shown. Note that the final maneuver shows abrupt changes
of the heading when the holonomic navigation is used. This
is because ARCO turns to get the desired heading when the
distance to the goal is lower than a threshold. The threshold
is 1.5 m in the tests performed.
One of the key features of the ARCO system is the
direct integration of people into the robots decision making
process. This way, the robot can act and decide considering
the presence of humans in the surrounding area.
People detection is particularly interesting in order to build
robot intelligence that treats persons differently, not just as
obstacles. This information can be used to build more efficient
solutions for safety and human acceptance.
Most of the approaches in the state of the art make use
of cameras and LIDAR for automatic people detection. While
these algorithms works well in general, they use to be limited
to detections in the sensor vicinity, and occlusions have a great
impact in the algorithm performance. In this sense, the use
of an ultra-wide-band (UWB) system provides a cost-efficient
and reliable localization system. Normally, this technology
cannot be used for people detection in the wild because it
requires the people to carry an UWB tag to be localized.
However, the environment can be much more controlled in
factories, so that workers can carry a tag, at least the ones
working in the warehouse.
Thus, ARCO has developed a module for range-only
localization, it was also integrated with a low-cost UWB
system. This software makes use of the individual UWB
range measurements and provides a position estimation of
the tags carried by the operators. Based on previous authors
works on range-only simultaneous localization and mapping
(RO-SLAM) [18] [19], this module implements an Extended
Kalman Filter (EKF) that integrates temporal constraints and
range measurements to estimate the position of the tag. This
module can be used to localize the robot and the people. While
filter updates are identical in both cases, different prediction
stages have been implemented in order to account for the robot
odometry or the human motion model respectively.
The experiments presented in Section VI shows that per-
sons can be localized with an averaged error of 20 cm
approximatelly when the area is well covered by the UWB
anchors. An interesting benefit of the developed approach
for localization is that it does not need three measurements
to estimate the tag position in 2D. Thanks to the temporal
constraint imposed by the filter, two range measurements are
enough to recover the tag pose.
The BPMN models described below have been designed
with Camunda Modeler, and then they have been integrated
into the MPMS application.
Camunda has a web application that can be accessed via
login by the system administrator or by the workers. The
application is employed to start and visualise the process
status, and perform manual tasks with forms.
The main process begins when the orders handling process
is started from the Camunda Tasklist (see Fig. 7). The orders
handling process runs on two paths in parallel: the first
one reads the email inbox folder checking whether there are
new incoming orders, and it stores them into a PostgreSQL
database. Then, one internal signal named ”email received”
is thrown. On the second path, a task reads the database for
unattended orders and if it finds any, it activates the warehouse
process. If there are no new order on the database, the token
stops until the ”email received” signal is caught. This signal
has the role to avoid unnecessary readings from the database.
Fig. 7: BPMN model: orders handling process.
The task ”Read emails” has an error catcher that allows the
system not to hang out if the email connection fails.
The warehouse process adds the use of lanes that represents
which actor is going to perform each task. The tasks performed
by the workers are manual tasks, which are performed through
forms that can be accessed via the Camunda web app.
The warehouse process starts establishing the connection
with the web server, then waits for the robot being ready (see
Fig. 8). Once ARCO confirms the availability, then the ”Move
To Material” task sends to the robot the coordinate where the
first raw material has to be picked up. One worker must go to
the goal position and load the robot with the correct amount of
raw material. Once the worker has loaded the material, it sends
a confirmation through a form and then the MPMS checks
if the weight of the raw material is correct. The robot then
moves to the mixing tank. The material is unloaded manually
and then, the robot goes to the next material location or, if all
the materials have been added to the mixing tank, the robot
goes back to the home position and the worker activates the
mixing procedure. Finally, the order status is updated in the
database and the connection with the web server is closed. The
warehouse process ends and the orders handling process keeps
querying the database for new incoming orders. If some error
is caught while running the process, the worker must access a
form and decide if the order must be cancelled, or completed
A testing environment that simulates the factory has been
implemented at the University (see Fig. 9). There are two
corridors and several places where to pick materials simulated
with boxes (A, B and C). During the experiments, people
will approach ARCO and some obstacles will appear in the
corridors in order to test the re-planning capabilities of the
Fig. 8: BPMN model: warehouse process.
Fig. 9: Places where to pick the boxes and mixing tank in the
testing environment considered.
The operating system used is Ubuntu 16.04. The code has
been written in C++ language and has been integrated in ROS
(Robot Operating System) framework. Particularly, with ROS
Kinetic Distribution.
Experiments carried out show the whole mission. First, an
order is received by the MPMS application. Then, the naviga-
tion goals are given one by one by the MPMS application to
ARCO. In the testing environment considered, ARCO should
reach A, B and C points to pick boxes and introduce each one
in the mixing tank. Therefore, the order of the points is: A,
tank, B, tank, C and tank.
A simplified BPMN model (Fig. 10) has been used in
these initial experiments, as the weighing system was not yet
available at that moment.
The tests performed evaluate the following characteristics
of the proposed system:
1) Full mission given by the MPMS application: ARCO
navigates efficiently and safely during the execution.
2) Detect and avoid: static obstacles (boxes) and people
approaching to the ARCO are considered. ARCO should
Fig. 10: BPMN model used during the experiments.
react correctly avoiding the obstacles or stopping when
obstacles are very close (the emergency stop system is
20 cm).
The system has been tested for hours in the testing envi-
ronment and all of them have been successful. A video of
one of the tests performed can be seen in here4. It shows
the reference system of ARCO (position of the robot), the
static map generated of the environment (dark pink), the global
costmap (it represents the inflated static map in light blue), the
local costmap (red square centered in the ARCO in red) and
the global path (green small square) and local paths (red small
Figure 11 depicts the robot trajectory of a performed mis-
sion (shown in the video). The Home position, wall, the places
where to pick the boxes and the tank are shown. The distance
travelled in this mission provided by the localization system
is 62.7m. The video shows a person approaching the robot
at the time 1:45 and the ARCO reacts avoiding them. This
manoeuvre can be seen in the segment Tank to B of the Fig. 11.
ARCO moves to the right side in order to avoid the detected
The UWB system has also been tested in real experiments.
In order to evaluate the precision of the UWB system, an
UWB tag was installed into the ARCO robot and the resulting
information was compared to the information from AMCL
laser based localization. Figure 12 shows the UWB and the
AMCL localization. Figure 13 shows the relative error of the
UWB localization with respect to the AMCL localization, it
can be seen how the average localization error is below 0.2m
and that the error in Y is larger than X, this is mainly produced
by the lack of trilateration in Y in this experimental setup. A
better distribution of the UWB anchors will definitely help to
reduce such errors.
The computation time of the Global planner depends on the
distance between the initial position and the goal. For example,
Fig. 11: ARCO trajectory in the mission recorded in the video
Full Mission 01.m4v.
Fig. 12: ARCO localization from AMCL localization (blue
line) and UWB estimation (red line).
Fig. 13: Relative error between UBW localization and AMCL
the computation time for distances greater than 10 m is 100-
120 ms approximately. For distances shorter than 10 m, the
global path can be computed in 40 ms approximately. On the
other hand, the computation time of the Local planner depends
on the number of obstacles detected. When ARCO moves in
straight line without obstacles, the mean computation time is
1.13 ms, the standard deviation is 1.18 ms and the maximum
computation time is 10 ms. In the cases where ARCO detects
obstacles and the Local planner should re-plan the global path
to avoid them, the mean computation time is 3.78 ms, the
standard deviation is 4.43 ms and the maximum computation
time in occasional cases is 42 ms. An important characteristic
of the implementation of the Local planner is that its execution
frequency is of 23Hz, so the this planner is very suitable for
the application considered in this paper as it ensures a quick
reaction when obstacles are detected.
Note that an emergency stop system is running at 40Hz,
this provides an emergency response. If an emergency stop
takes place, ARCO will move again when the distance to the
obstacle or person is greater than 0.6 m.
In this paper, a human-robot co-working system for ware-
house automation has been presented. The system uses an
automated guided vehicle (AGV) called ARCO and a Man-
ufacturing Process Management System (MPMS) controls all
the process.
The proposed system is able to carry a complete mission,
navigating efficiently and safely into the testing environment.
The system has been tested in a real environment. The duration
of all the tests and the absence of errors in the performed tests
validate the system.
The MPMS achieved the goal of controlling several pro-
cesses at the same time, on an heterogeneous ensemble of
connected devices. The order handling and the robot naviga-
tion were successfully managed, and the process execution was
easily monitored in real time through the cockpit. Camunda
BPM turned out being an effective tool for implementing
processes into a connected industry 4.0 environment.
Future work will consider the integration of a weighting
system for the raw materials. Also, kinodynamic constraints
will be considered into the patch tracker or the planner so that
ARCO can act considering the mass of the system.
[1] E. Commission, “Europe 2020: A strategy for smart, sustainable and
inclusive growth,” Brussels, Belgium, 2010.
[2] J. P. Holdren, T. Power, G. Tassey, A. Ratcliff, and L. Christodoulou,
“A national strategic plan for advanced manufacturing,” URL: http://, National Institute of Standards and
Technology, Tech. Rep., 2012.
[3] J. Zhou, “Intelligent manufacturing-main direction of made in china
2025,” in China Mech. Eng., vol. 26, no. 17, 2015, pp. 2273–2284.
[4] I. F. of Robotics, “Robots double worldwide by 2020,
2020, 2018.
[5] H. Lasi, P. Fettke, T. Feld, and M. Hoffmann, “Industry 4.0,” in Business
& Information Systems Engineering, vol. 6, no. 4, 2014, pp. 239–242.
[6] F. D. P. R. Wurman and M. Mountz, “Coordinating hundreds of
cooperative, autonomous vehicles in warehouses,AI Magazine, vol. 29,
no. 1, pp. 9–20, 2008.
[7] K. C. T. Vivaldini, J. P. M. Galdames, T. S. Bueno, R. C. Araujo,
R. M. Sobral, M. Becker, and G. A. P. Caurin, “Robotic forklifts for
intelligent warehouses: Routing, path planning, and auto-localization,”
in Proceedings of the 2010 IEEE International Conference on Industrial
Technology, Vi˜
na del Mar, Chile, 14-17 March 2010, pp. 1463–1468.
[8] I. F. A. Vis, “Survey of research in the design and control of automated
guided vehicle systems,” European Journal of Operational Research,
vol. 170, pp. 677–709, 2006.
[9] N. Pinkam, F. Bonnet, and N. Y. Chong, “Robot collaboration in ware-
house,” in In Proc. ACCProceeding of the 16th International Conference
on Control, Automation and Systems (ICCAS 2016), Gyeongju, Korea,
16-19 October 2016, pp. 269–272.
[10] H. Martinez-Barber and D. Herrero-Prez, “Autonomous navigation of
an automated guided vehicle in industrial environments,Robotics and
Computer-Integrated Manufacturing, vol. 26, pp. 296–311, 2010.
[11] S. Zor, F. Leymann, and D. Schumm, “A proposal of bpmn extensions for
the manufacturing domain,” in Proceeding of the 13th International Joint
Conference on Computer Science and Software Engineering (JCSSE),
Khon Kaen (Thailand), 13-15 July 2016.
[12] R. Petrasch and R. Hentschke, “Process modeling for industry 4.0
applications. towards an industry 4.0 process modeling language and
method,” in Proceeding of the 13th International Joint Conference
on Computer Science and Software Engineering (JCSSE), Khon Kaen
(Thailand), 13-15 July 2016.
[13] M. Polderdijk, I. Vanderfeesten, J. Erasmus, K. Traganos, T. Bosch,
G. van Rhijn, and D. Fahland, “A visualization of human physical
risks in manufacturing processes using bpmn,” in In Business Process
Management Workshops - BPM 2017 International Workshops, 2018,
pp. 732–743.
[14] I. V. P. Grefen and G. Boultadakis, “D2.2a - complete system design,”
URL: D2.
2%20(Public%20Version).pdf, Eindhoven University of Technology and
European Dynamics, Tech. Rep., 2015.
[15] C. S. Giorgio Grisetti and W. Burgard, “Improved techniques for grid
mapping with rao-blackwellized particle filters,” IEEE Transactions on
Robotics, vol. 23, pp. 34–46, 2007.
[16] W. B. S. Thurn and D. Fox, Probabilistic Robotics. The MIT Press,
[17] A. Nash, S. Koenig, and C. Tovey, “Lazy theta*: Any-angle path
planning and path length analysis in 3d,” in Proceedings of the AAAI
Conference on Artificial Intelligence (AAAI), Atlanta, Georgia, USA,
11-15 July 2010, pp. 147–154.
[18] F. Fabresse, F. Caballero, and A. O. I. Maza, “An efficient approach for
undelayed range-only slam based on gaussian mixtures expectation,”
Robotics and Autonomous Systems, vol. 104, pp. 40–55, 2018.
[19] F. Caballero, L. Merino, and A. Ollero, “A general gaussian-mixture
approach for range-only mapping using multiple hypotheses,” in IEEE
International Conference on Robotics and Automation (ICRA), Anchor-
age, AK, USA, 3-7 May 2010, pp. 4404–4409.
... On the other hand, managers could even improve the well-being of their operators through the reduced physical load. Some authors (Papcun et al. 2019;Rey et al. 2019) comment that reducing the physical load on the order pickers could increase safety and improves the economic results. Zhang, Winkelhaus, and Grosse (2021) also commented that order pickers working in close collaboration with a PTR have less physical load. ...
... Additionally, it poses localization limitations and might not adapt to the environment suitably [8] [9]. Even with advanced robotic coworker models such as ARCO [10], several Kino dynamic constraints have not been optimized yet, affecting the tradeoff further. In the ease of using the control of discrete-event systems for multi-robot warehouse automation [11], several other factors come into play that triggers more disadvantageous aspects. ...
Conference Paper
The COVID-19 pandemic has revolutionized the world and set forth a new normal. The sharp decline of availability of human resources has adversely affected industries relying on laborious work. Hence, these industries are experiencing a halt in their manufacturing processes, leading to a loss of time and business opportunities. In consultation with one such industry looking to automate their material handling system, we present a survey of possible automation solutions that could reduce human intervention to increase production efficiency based on certain key parameters. Automating the movement of machined parts from one station to another within a manufacturing plant is essential since manual part handling often leads to part mix-up, part damage and low traceability of components.
... • Freight transport [15]: thanks to the autonomous navigation system, the robot can fully assist people in activities such as transport of various raw materials, storage and delivery. ...
... As a consequence of the pandemic period also, the e-commerce sector is growing fast and the only way to supply the chain is to go towards smart and autonomous warehouses [3]. For these reasons, robotic applications are growing in daily operations inside warehouses [4][5][6]. Several companies are already using this technology to speed up their operations while many others are on their way to incorporating it. However, to increase the autonomy level in these GPS-denied environments, it is first required to solve the problem of localizing the robot with adequate accuracy. ...
Full-text available
For the commercial sector, warehouses are becoming increasingly vital. Constant efforts are in progress to increase the efficiency of these facilities while reducing costs. The inventory part of the goods is a time-consuming task that impacts the company’s revenue. This article presents an analysis of the performance of a state-of-the-art, visual-inertial odometry algorithm, SVO Pro Open, when varying the resolution and frequency of video streaming in an industrial environment. To perform efficiently this task, achieving an optimal system in terms of localization accuracy, robustness, and computational cost is necessary. Different resolutions are selected with a constant aspect ratio, and an accurate calibration for each resolution configuration is performed. A stable operating point in terms of robustness, accuracy of localization, and CPU utilization is found and the trends obtained are studied. To keep the system robust against sudden divergence, the feature loss factor extracted from optical sensors is analyzed. Innovative trends and translation errors on the order of a few tens of centimeters are achieved, allowing the system to navigate safely in the warehouse. The best result is obtained at a resolution of 636 × 600 px, where the localization errors (x, y, and z) are all under 0.25 m. In addition, the CPU (Central Processing Unit) usage of the onboard computer is kept below 60%, remaining usable for other relevant onboard processing tasks.
... The proposed tool running externally from the robot can use BPMN models to abstract the implementation of one single robot, and to generate and execute the related mission. Another integration of the BPMN standard with an autonomous robot is presented in [16]. The authors develop and automatize a warehouse process by implementing a human-robot cooperation system managed by a web interface able to control the entire process in real time. ...
Programming robotic systems is often a challenging task requiring advanced skills, especially when the goal is to ensure loosely-coupled coordination in heterogeneous Multi-Robot Systems (MRSs). Model-driven approaches for robotic system engineering have shown their benefits in facilitating the development of robots’ behavior, controllers, and system components. However, the state of the art still lacks contributions addressing crucial aspects of the model-driven approach applied to MRSs, such as developing robots’ distributed cooperation through models supporting the communication among robots. In this paper, we present a novel framework for modeling, configuring and enacting the cooperative behaviors of MRSs through collaboration diagrams as provided by the BPMN 2.0 standard. The advantages of our solution lie, indeed, in the use of BPMN, which provides easily understandable and highly expressive diagrams for representing the cooperation among distributed robots, and benefits from a wide list of supporting tools. Starting from the selection of BPMNelements, we define a set of guidelines for driving the developer in modeling an MRS mission using BPMN. The developer configures the resulting collaboration diagram to link elements in the model to the robotic middleware, ROS2 in the toolchain we implemented. Finally, the configured model is enacted by BPMN engines integrated into the ROS2 middleware run by each robot involved in the MRS, thus obtaining a fully distributed cooperation. We assess our framework’s effectiveness through experiments in simulated and real environments.
... or enabling production flows), either production robots or autonomous mobile robots (AMR) can be used [3], turning factories into an unstructured environment with manufacturing systems and routing of goods changing dynamically [4]. automated guided vehicle (AGV) is an example of efficient warehouse systems where humans are either replaced by robots or collaborate closely with them [5]. The use cases above set new requirements for communication technologies with higher throughput, lower latency, and higher reliability than current wireless systems can offer [6], [7], [8]. ...
Full-text available
High throughput, low latency, and high reliability in proximity communications for swarm robotics can be achieved using decentralized cooperative resource allocation schemes. These cooperative schemes minimize the occurrence of half-duplex problems, reduce interference, and allow a significant increase in the achievable swarm density, but requires additional signaling overhead, which makes them potentially more prone to performance degradation under realistic operation conditions. These conditions include both data, signaling, and their interdependence evaluated jointly. The negative impact of the signaling errors requires incorporating enhancement techniques to realize the full potential of the cooperative schemes. Particularly, in this paper and for this purpose, we evaluate the effects of hybrid automatic repeat request (HARQ), link adaptation by aggregation (LAAG) and beam selection by using directional antennas in the cooperative schemes, and compare performance with 3rd Generation Partnership Project (3GPP) NR sidelink mode 2 (including signaling) using the same techniques. Additionally, we include a comparison of the required number of control signals between sidelink mode 2 inter-UE coordination (IUC) and cooperative schemes, and introduce a decentralized rebel sub-mode behavior in our group scheduling scheme to further improve the performance at the 99.99 percentile. The simultaneous use of all these enhancement techniques in our cooperative schemes considerably reduces the impact of signaling errors and thereby increases the supported swarm size compared to sidelink mode 2.
This is the conference proceedings of International Conference on Intelligent Systems and Applications (ICISA-2022)
In the past couple of decades, the applications of industrial robotics in the industries have seen exponential growth due to the advancements in technologies. The concept of Industry 4.0 is the next generation manufacturing system which is a confluence of smart sensors, electric and electronics technologies, and core manufacturing technologies. Warehouse automation plays a critical role to meet today’s demanding goals of supply chains, and warehouse automation technologies are required to be unprecedentedly efficient, adaptive, scalable, and fault-tolerant. This paper presents an application of AI-based smart robotics system for intelligence warehouse automation. In the current work, the territory of the warehouse is explored by the robots. AI-based path planning algorithms, viz. rapidly exploring random trees (RRT), RRT*, A* are applied for the path planning of robots in the warehouse. The results show the potential application of robotics systems in the domain of intelligent automation of warehouses.
Artificial intelligence is an emerging field that is undergoing humongous development. Self-driving technology is a major part of that development. Various companies like Tesla, Waymo, Ford, and Audi are part of research and development in this field. The modern self-driving vehicles that work on Level 5 autonomy utilize lidar technology along with computer vision frameworks to get a complete autonomous driving experience. These companies are based in the USA or well-developed countries. Their implementation in India, which is a developing country, is to be determined. To get a fair idea of how these vehicles and their technology will perform in the Indian scenario, a testing plan is devised in this work. In this paper, a YOLO object detection model was custom trained on images taken in the Indian traffic scenario and compared with a pre-trained model which has been trained on COCO dataset which contains traffic images from various foreign countries like the USA and Europe. The comparison has been done on the metrics like accuracy in detection of classes, precision, recall, and F1-score.
Humans still need to access the same area as automated systems, like in warehouses, if full automation is not feasible or economical. In such shared areas, critical interactions are inevitable. The automation of vehicles is usually tied to an argument on improved safety. However, current standards still rely also on the awareness of humans to avoid collisions. Along with this, modern intelligent warehouses are equipped with additional sensors that can help to automate safety. Blind corners, where the view is obscured, are particularly critical and, moreover, their location can change when goods are moved. Therefor, we generalize a concept for safe interactions at known blind corners to movements in the entire warehouse. We propose an architecture that uses infrastructure sensors to prevent human-robot collisions with respect to automated forklifts as instances of driverless industrial trucks. This includes a safety critical function using wireless communication, which sporadically might be unavailable or disturbed. Therefore, the proposed architecture is able to mitigate these faults and gracefully degrades the system’s performance if required. Within our extensive evaluation, we simulate varying warehouse settings to verify our approach and to estimate the impact on an automated forklift’s performance.KeywordsDriverless industrial trucksHuman-robot interactionInfrastructure sensorsWarehouse
Conference Paper
Full-text available
The term Industry 4.0 derives from the new (fourth) industrial revolution enabling suppliers and manufacturers to leverage new technological concepts like Internet of Things, Big Data, and Cloud Computing: New or enhanced products and services can be created, costs be reduced, and productivity be increased. Similar terms are Smart Factory or Smart Manufacturing. The ideas, concepts, and technologies are not hyped anymore-they have at least partly become a reality, but many software specifications and development aspects are still not sufficiently covered, e.g., standardization, specification, and modeling languages. This paper presents an Industry 4.0 process modeling language (I4PML) that is an extension (UML profile with stereotypes) of OMG's BPMN (Business Process Model and Notation) standard. We also describe a method for the specification of Industry 4.0 applications using UML and I4PML.
Full-text available
This paper deals with range-only simultaneous localization and mapping (RO-SLAM), which is of particular interest in aerial robotics where low-weight range-only devices can provide a complementary continuous estimation between robot and landmarks when using radio-based sensors. Range-only sensors work at greater distances when compared to other commonly used sensors in aerial robotics and they are low-cost. However, the spherical shell uniform distribution inherent to range-only observations poses significant technological challenges, restricting the approaches that can be used to solve this problem. This paper presents an undelayed multi-hypothesis Extended Kalman Filter (EKF) approach based on Gaussian Mixture Models (GMM) and a reduced parameterization of the state vector to improve its efficiency. The paper also proposes a new robot-to-landmark and landmark-to-landmark range-only observation model for EKF-SLAM which takes advantage of the reduced parameterization. Finally, a new scheme is proposed for updating hypothesis weights based on an independence of beacon parameters. The method is firstly validated with simulations comparing the results with other state-of-the-art methods and later validated with real experiments for 3D RO-SLAM using several radio-based range-only sensors and an aerial robot.
Grids with blocked and unblocked cells are often used to represent continuous 2D and 3D environments in robotics and video games. The shortest paths formed by the edges of 8-neighbor 2D grids can be up to 8% longer than the shortest paths in the continuous environment. Theta* typically finds much shorter paths than that by propagating information along graph edges (to achieve short runtimes) without constraining paths to be formed by graph edges (to find short "any-angle" paths). We show in this paper that the shortest paths formed by the edges of 26-neighbor 3D grids can be 13% longer than the shortest paths in the continuous environment, which highlights the need for smart path planning algorithms in 3D. Theta* can be applied to 3D grids in a straight-forward manner, but it performs a line-of-sight check for each unexpanded visible neighbor of each expanded vertex and thus it performs many more line-of-sight checks per expanded vertex on a 26-neighbor 3D grid than on an 8-neighbor 2D grid. We therefore introduce Lazy Theta*, a variant of Theta* which uses lazy evaluation to perform only one line-of-sight check per expanded vertex (but with slightly more expanded vertices). We show experimentally that Lazy Theta* finds paths faster than Theta* on 26-neighbor 3D grids, with one order of magnitude fewer line-of-sight checks and without an increase in path length.
How to transform from a big nation in manufacturing to a powerful one is a great subject of China's economic development in the new era. This paper discussed the guiding ideology, strategic deployment, basic policy and strategic measure of “made in China 2025” based on detailed analyses of current manufacturing situations of China, and indicated that “Internet+advanced manufacturing+modern service industry” will become a new engine of China's economic development. It is pointed out that digitization, networking and intellectualization of manufacturing industries are the core technologies of a new round industrial revolution, and should be used as commanding point, breakthrough and main direction of “made in China 2025”. An overall thorough interpretation of above points was put forward. ©, 2015, China Mechanical Engineering Magazine Office. All right reserved.
The manufacturing domain can be characterized by the concepts of product, process and resource. The coordination of humans, machines and materials is needed to attain a desired product by using knowledge, time, money and energy optimally. Business processes define such coordination. Business Process Model and Notation (BPMN) is an industry standard for modeling business processes. This standard is not being applied to the manufacturing domain so far. In this paper, we propose BPMN extensions to support process modeling in the manufacturing domain, especially the modeling of production processes. We motivate and justify our proposed extensions by an exemplary assembly process.
The research presented in this paper approaches the issue of navigation using an automated guided vehicle (AGV) in industrial environments. The work describes the navigation system of a flexible AGV intended for operation in partially structured warehouses and with frequent changes in the floor plant layout. This is achieved by incorporating a high degree of on-board autonomy and by decreasing the amount of manual work required by the operator when establishing the a priori knowledge of the environment. The AGV's autonomy consists of the set of automatic tasks, such as planner, perception, path planning and path tracking, that the industrial vehicle must perform to accomplish the task required by the operator. The integration of these techniques has been tested in a real AGV working on an industrial warehouse environment.