Intelligent Service Robotics (2021) 14:313–325

https://doi.org/10.1007/s11370-021-00363-w

ORIGINAL RESEARCH PAPER

Coordination of multi‑robot path planning for warehouse application

using smart approach for identifying destinations
Kaushlendra Sharma1 · Rajesh Doriya1

Received: 1 September 2020 / Accepted: 3 March 2021 / Published online: 27 March 2021
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021

Abstract
Path planning and coordination in a multi-robot system are important and complex tasks in any environment. In a multi-
robot system, there can be multiple objectives to be achieved by multiple robots simultaneously. Nowadays, many mobile
service robots are being used in warehouses to reduce running costs and overheads. In a large warehouse, there can be mul-
tiple robots to handle the number of operations. Planning a path means to find out the optimal route, and coordinating them
means a collision-free route. To get both the parameters to reach their optimal level becomes a tedious task to achieve. The
efficiency of overall warehouse operation can be improved by adequately addressing the coordination and path planning
issues among the robots. In warehouses, each robot has to navigate to its destination by finding a collision-free optimal route
in coordination with other robots. In this paper, a comparative study with the acclaimed path planning and coordination has
been presented. The proposed smart approach has been presented for a multi-robot system to find a collision-free optimal
path in a warehouse to handle storage pods. This paper proposes a smart distance metric-based approach for a multi-robot
system to identify their goals smartly and traverse only a minimal path to reach their goal without getting being collided. It
uses a smart distance metric-based approach to find the intended path. The proposed work performs better when compared
with other works like A* and ILP. It is strictly monitored that there is no collision occurred during execution. Three different
instances of a warehouse have been considered to carry out the experiments with parameters such as path length, average
path and elapsed time. The experiments with 800 pods and 16 robots report the improvement in performance up to 2.5% and
13% in average path length and elapsed time.

Keywords Multi-robot · Path planning · Coordination · Warehouse

1 Introduction and related work and some of them inherited those concepts into multi-robot
systems as well. However, the moment when path planning
Multi-robot system is required to coordinate well with is applied to the multi-robot system, there exists one more
each other to carry out a challenging and complex task like issue which needs to addresses thoroughly, which is known
managing the goods and services in the warehouse [1–3]. as coordination. There are also some work present which has
Achieving more than one objective at the same instance and been proposed by the researchers regarding the coordination
in the shared workspace makes it an uphill task to achieve. scheme [4]. Combining these two issues parallel requires a
Multiple robots require two crucial parameters to get opti- smart approach to achieve synchronization between these
mized up to their maximum in such cases. The first one is two parameters. Here, in this section, we present a detailed
to derive an optimal path for each robot, and the second definition of all the essential aspects related to multi-robot
is to maintain coordination among them. A considerable path planning and coordination. Also, this section covers
number of work has been done in the area of path planning some notable contribution made and a literature survey done
for robots, though most of them focus on a single robot, in this regard.

1.1 Path planning
* Kaushlendra Sharma
[email protected] Due to the working of the path planning algorithm and its
1 behavioural complexity, it is considered as the NP-hard
National Institute of Technology Raipur, Raipur, India

13
Vol.:(0123456789)
314 Intelligent Service Robotics (2021) 14:313–325

problem [5]. Initially, the path planning was done for the present in the workspace. Thus any conventional path plan-
robotics arm only for some manufacturing purposes in ning algorithm has to behave differently for finding an opti-
industries. Later on, with the wide acceptability of robotics mal path for each robot due to the continuous change in the
in multiple sectors, different types of robots were brought environment. The second thing which makes it complicated
into use for serving the task, which leads to an increase in is that it is required in the working architecture to imbibe a
the work related to path planning algorithms. There exist a coordination mechanism to manage all the robots working in
long history of path planning algorithms. Starting from the that workspace [17]. Thus, while dealing with the problem
Heuristic approach to solve the path planning problem to the of the multi-robot system, it becomes mandatory to deal with
evolutionary and hybrid algorithm to find an optimal path the two issues simultaneously, the first one being the path
from source to destination, while the robot tends to traverse planning algorithm to find an optimal path for each robot as
in the given configuration space. The definition of path plan- well as ensuring the optimal average path of all the robots
ning is to plan a collision-free route from source to destina- working in the shared workspace. The second issue to deal
tion, while the goal of different path planning algorithm is to with is having an effective coordination approach such that
make that path optimum and safe. Several new technologies it makes all the robots to move on their optimal path and that
were popping up with each passing days. A good number of too, without any collision.
optimization techniques and algorithms has been on the top
for a few years. Several researchers have consumed those 1.3 Coordination
concepts for solving the path planning problem for robots.
Some of the latest entrants in the queue are, memetic algo- Coordination is required when there is more than one task to
rithm [6], genetic and adaptive fuzzy control [7], delayed instate. The multi-robot scenario is one among it [18]. The
PSO algorithm [8], beetle antennae search algorithm [9], basic principle of coordination is to avoid collision among
Smoothed A* [10]. There are also many advanced versions the robots while traversing, and after doing so, the second
of the A*, RRT, PRM algorithm proposed by the researchers most crucial thing to manage is whether they are central-
to solve the path planning problem in multiple domains [11] ized or decentralized coordinated [19]. In a centralized
[12]. These algorithms can also be considered as the core scheme [20], there is a supervisor robot which synthesis all
algorithms for solving robot path planning. Path planning the robots and performs all the required calculation for safe
again can be studied in terms of its environmental set-up execution [21]. While in a decentralized scheme [22, 23] all
like static, dynamic, two-dimensional, three-dimensional the robots perform individually and distribute their informa-
and many other ways. To work in a dynamic environment, tion to other robots [24]. Comparing it with the centralized
path planning is done by considering virtual static obstacle scheme, it is much more flexible and reliable in adapting to
to avoid collisions. Each time the path planning is done con- the environment but is more expensive in terms of the cost it
sidering the time parameter, the updated virtual position of incurs. There are also some applications that uses the hybrid
an obstacle is considered. Though that fall separately into approach [25, 26] by combining both the centralized and
other zones, that’s why we do not include the discussion of decentralized methods of coordination. There are multiple
that part. techniques [27, 28] proposed by the researchers to coordi-
nate the multiple robots in the given workspace. There are
1.2 Multi‑robot system some more additional issues which are required to handle
and manage simultaneously. The issues like how the robots
A multi-robot system is used to accomplish a complex task. are made to traverse by keeping a minimal distance with
A single robot is capable of handling most of the simplified other robots as increasing the distance between them will
jobs. Multi-robot is required when multiple objectives need affect their optimal paths. In the same way, it is also required
to be fulfilled simultaneously, which makes it complicated. to take care of the speed of each robot with which they are
Application like dynamic mission planning [13], collective moving. It should also be maintained at its optimal level, as
construction [14], multiple task allocation [15], mapping increasing or decreasing from their optimal speed will lead
environmental variables [16] and many other applications to safety concerns and may make the whole workspace an
have used multiple robots in the past. It requires an entirely unsafe environment.
different architecture to work upon. Each robot is assigned
a unique goal to satisfy; every single robot has to traverse an 1.4 Related work
optimal path for reaching their respective destination from
the start point. Managing multiple robots simultaneously in Here, in this subsection, we have analyzed some of the
a shared workspace makes it more dynamic and complex. prominent work in the area of path planning and coordina-
Dynamic in the sense that each other robot is an obstacle tion. A short discussion is presented below, which helps in
for every other robot, apart from the static obstacles already indulging into the core working of a multi-robot system.

13
Intelligent Service Robotics (2021) 14:313–325 315

Xinye et al. in the [29] paper presents the problem of The extant push and spin (PASp) algorithm ensure provid-
path planning for a group of mobile robots with multiple ing a complete solution to the multi-robot path planning
targets as multi-travelling salesman problem (MTSP) with problem. An improved push and spin (PASp+) algorithm
one or more depots. A bi-objective ant colony optimization are devised to obtain an optimal path using the standard
(ACO) algorithmic approach, which is based on the memetic PASp algorithm two ways were adapted, which include the
algorithm, is presented to solve the problem. So the essen- use of smooth operation that eliminates redundant moves in
tial requirement is to ensure that every robot must visit at the path, and the other involves usage of heuristic search to
least one target, and each target is visited once by one of the explore other available paths. The PASp algorithm involves
robots. In addition to this, the simultaneous optimization of the usage of push and spin operations when two robots are
the total path length and maximum path length of the robots detected simultaneously at a common vertex position along
are to be achieved. The approach employees ACO based on their paths. Instead of spin operation, a heuristic value com-
a memetic algorithm that uses local search and integrates it parison with other available paths is done, and the smooth
with sequential neighborhood descent for optimization [30]. operation is applied at the end on the list of solutions
The experimental results of this approach tested in a static obtained. The proposed approach was tested and compared
environment are compared with other classical algorithms, with the standard algorithms, which shows that the proposed
which show that the proposed approach gives better results. approach is better in performance.
Rami et al. paper [31] describes an approach based on Reducing down the related work, this paper focuses on
probabilistic neuro-fuzzy logic. This approach involves two the warehouse applications implementing a multi-robot sys-
layers of fuzzy architectures and based on the leader-fol- tem. In context to that a small survey is presented in Table 1.
lower scenario. First is the probabilistic fuzzy control that Warehouse application is one of the areas which is best uti-
takes care of uncertainties and errors to avoid disturbance in lizing the multi-robot system to serve and fulfil the require-
path traversal and approximates the position of the robot rel- ments. It is again found that most of the solution proposed
atively. The neuro-fuzzy Inference system layer is designed is centralized or distributed in nature.
to establish a leader-follower movement among the robots.
The problem formulation is depicted using fuzzy architec-
ture diagrams. The neuro-fuzzy inference system acts as a 2 Problem definition
navigation controller for each robot. Its input coordinates
and the orientation of the robot, as well as the output, are The problem is set up in a static warehouse environment
the linear speed and angular speed of the robot. The experi- with multiple robots whose initial configurations are defined,
mental results of this approach are obtained by simulation on and a set of goal positions are given. So the problem can
different environments; the leader and follower movements here be viewed as optimal target assignment to the robots
are observed in the experiment. The future scope involves followed by path planning to develop a collision-free path
testing the current approach in a real complex environment for each robot to reach the target. In a multi-robot system,
and to develop a technique to improve the performance and path planning requires coordination to obtain a feasible
to reduce the complexity. collision-free path in minimum moves for the robot. The
Jianjun et al. in the paper [32] mainly focuses on improv- static warehouse environment here consists of static obsta-
ing the path planning for a multi-robot system using the cles, which are a non-traversable region for the robots, and
genetic algorithm (GA), which is achieved by using a the remaining free space region is traversable are shown in
memetic approach. The proposed improved memetic Fig. 1 The necessary information required for path planning
algorithm-based (IMA) approach involves in implement- involves target position to be reached, so target assignment
ing GA with variable length chromosome using the two- plays a key role here.
point crossover and bacterial mutation operations, which In general, the multi-robot path planning system has a
avoids the optimum local problem. A search strategy that defined mapping of a target for each robot to reach the target.
combines local neighbors search with disorder strategy is So target assignment is an additional task that effects the
used to improves the overall convergence rate. Further- final path obtained. The primary objective in target assign-
more, an approach to deal with multi-robot path planning ment is to assign targets such that future paths obtained
in a dynamic environment is also proposed. The experi- have minimum path length and computation time. Using
mental results obtained by real-time testing and simulations random permutations for target assignment result in con-
are specified that shows the significant improvement in the siderable path length and computation time. So a proper
results in comparison with the general genetic and memetic approach with prior consideration of path lengths as a factor
approach. is desirable.
Ebtehal et al. paper [33] aims at developing a complete Consider a set of n robots in a multi-robot system where
and optimal solution for multi-robot path planning problem. each robot is represented as Ri where i =1, 2,..., n indicates

13
316 Intelligent Service Robotics (2021) 14:313–325

Table 1  A brief survey on Coord.

the multi-robot path planning Tech.
algorithms and coordination
techniques for different Author Path Planning C D Application Parameters Year R.I. Ref
applications
Warren et al. Artificial PF ✓ No Static Envi. Path length 1990 281 [34]
Svestka et al. PRM Algo No ✓ Car Parking Time Factor 1995 209 [35]
Petr et al. PRM Algo ✓ No Robot Car Comp. Time 1998 303 [36]
Burgard et al. Probabilistic ✓ No Unkown Envi Target Sele 2000 932 [37]
Bennewitz et al. Hill Climbing ✓ No Real Robots Path Length 2001 194 [38]
Sanchez et al. PRM Planner ✓ No Welding Station Completeness 2002 188 [39]
Kalra et al. Hoplites Algo No ✓ Mult Agent Plan Coordination 2005 179 [40]
Wurm et al. Segmentation No ✓ Unknown Envi. Exp.Time 2008 248 [41]
Hollinger et al. Approximation No ✓ Indoor Envi Capture Time 2009 159 [42]
Van et al. Decoupling ✓ No Unknown Running Time 2009 145 [43]
Hollinger et al. Online Algo. ✓ No Simulation Comp. Cost 2010 86 [44]
Ding et al. Optimaility ✓ No UAV Convoy Path Time 2010 127 [45]
Wagner et al. M* StarAlgo No ✓ Warehouse Min Cost Path 2011 149 [46]
Michael et al. Any-Com Algo No ✓ Dynamic Teams Completeness 2014 17 [47]
Contreras et al. ABC-EP Algo ✓ No 2D-Map Path Length 2015 140 [48]
Huijiao et al. Auction Algo No ✓ Dynamic Envi. Task Alloc. 2015 08 [49]
Glenn et al. M* Star Algo ✓ No Subdimensional Min Cost Path 2015 160 [50]
Claes et al. Monte Carlo No ✓ Static Envir. Planning Time 2017 28 [51]
Araki et al. Priority Plan No ✓ Swarm Robots Motion Control 2017 29 [52]
Bolu et al. Modified A* No ✓ Smart Warehouse Completion Time 2019 02 [53]
Kumar et al. GA and A* ✓ No Warehouse Order Picking 2018 15 [54]
Han et al. Heuristic No ✓ Warehouse Env. Comp. Time 2019 05 [30]

the robot id. A matrix denoting mapping between robots

and goals D. Configuration of robot Ri at time t is given as
per the Eq. (1).
Ci (t) = (xi , yi , t) (1)
Now, the collision between two robots Ri and Rj is defined
as per Eq. (2).
Collision(i, j) → ∃t ∶ Ci (t) = Cj (t) where i ≠ j
(2)
i.e, → ∃t ∶ (xi , yi , t) = (xi , yi , t) where i ≠ j

To obtain globally optimal paths, we define a parameter

average path length( 𝜇pl ) mathematically in 3.
n
∑
𝜇pl = pij (3)
i=1,j=1

where n is the number of robots in the system and pij denotes

the length of the path traced from Robot R(i) to the goal Gj .
Fig. 1  A sample warehouse map where the black cell represents the The objective is to determine a collision-free and path with
storage place and are also the destination for robots, whereas the minimal path length for each robot in the system to reach
white cell represents the free space where traversal is possible the goal state. These two conditions are to be satisfied to

13
Intelligent Service Robotics (2021) 14:313–325 317

ensure the solution is optimal, which can be mathematically the positions of all the robots and destinations in the work-
represented, as mentioned in the following equation. Firstly, space. The block adjacent to it is showing that all the robots
collision-free implies 4. present in the workspace are connected to the destination.
After that, their respective distance is calculated. Then in
Collision (i, k) = 𝜙 ∀ Ri , Rk where, i ≠ k (4) the next stage, the distance of all the robots from all the
Which states that all the pairs of robots whose paths involved available destination is calculated. After that, the pairing
in a collision should be empty. Secondly, 𝜇pl should be mini- is done, which means each robot is selected for a specific
mum. Consider the set of optimal paths (S) to all the goals destination to traverse. This is done by selecting the mini-
in the system is given by the following 5 mum distance of each robot from their destination. Figure 2v
shows the major limitation of this existing approach which
S = [Pij,... ] ∀i, j where Di,j = 1 (5) states that while calculating the distance, it does not consider
the obstacle present in between the robot and destination.
And, min (𝜇pl ) and collision (i, k) = 𝜙 for each pair pair of In Fig. 3, the improvised method of solving this approach
Ri , Rk robots in the system is depicted. Here, also subfigures of Fig. 3i–iv depicts the
different instances of the proposed approach, and subfigure
Fig. 3v shows the solution of the proposed approach. For
3 Proposed solution calculating the distance of each robot from the destination,
intersection points have been taken into consideration. Inter-
This section presents the description of the proposed path section points are the position of two points of the flip side
planning method for the multi-robot environment and the of the obstacle, which helps in calculating the actual distance
process of their coordination. Here in this paper, we intend considering the presence of obstacle lying in between the
to propose an improved smart selection of pairs to find the robot and destination. The distance ratio helps in finding the
nearest best possible destination for each Robot Ri. This destination for each robot is shown in Fig. 6. This ratio helps
improved smart selection has an advantage over primitive in finding the optimum distance to be travelled by overcom-
selection as primitive selection uses the traditional distance ing the first selection, which does not consider the presence
metrics, which does not consider obstacles while the former of an obstacle while calculating the distance. The distance
one will consider obstacles. Figures 2 and 3 represent the ratio 𝛼 helps in finding the collision-free optimal path. The
pictorial description of the robot’s smart selection mecha- proposed concept is then applied to solve the problem of
nism. Subfigures (i–iv) of Fig. 2 represents the incremen- warehouse application having multiple robots sharing the
tal instances of the existing problem. Figure 2i is showing same workspace and having multiple destinations.

Fig. 2  Different instances of the execution and representing the existing problem of the system

13
318 Intelligent Service Robotics (2021) 14:313–325

Fig. 3  Different instances of the execution and showing the distance calculation DRO and DOG and the calculation of distance ration 𝛼 of the pro-
posed approach

3.1 Proposed path planning method estimating the cost of the path. For each goal from the avail-
able robots, this cost is obtained by summation of total path
The key steps involved in the solution is first to convert the length and collisions across the planned path. The goal is
environment map into binary form. Then, the second step assigned to the robot with minimum cost, and the process
involves the generation of source–destination pairs using is repeated for all the goals in the system. A scenario that
the distance ratio for target assignment. The following step elucidates the target assignment is shown in Figs. 2 and 3.
involves the usage of the distance metric approach for path
planning ensuring collision-free paths are obtained. Here,
the mutual collisions among robots are handled by colli-
sion avoidance through the reservation table. Initially, the
multi-robot environment is generated. A reference map is
given as the input to find the path, which is a static ware-
house environment map. On conversion, the image obtained
contains the black pixels indicate the storage pods that are
non-traversable and white pixels denoting traversable region.
The start positions are defined for each robot, and the smart
selection algorithm is used to assign goals or targets that
improve path planning in terms of path length. A goal map-
ping matrix ′ D is obtained as a result of smart selection per-
formed that is used for path planning further. According to
the smart selection approach, for every goal Gi we will be
having many distance ratios 𝛼i . The pair having the small-
est value of 𝛼 will be selected, as the selected robot will be
having maximum distance from the obstacle and minimum 3.2 Proposed coordination method
from the goal position. Thus, it gives a better estimation of
free space to be traversed in the path. The process of target To avoid robots colliding with other robots and also from the
assignment in a static warehouse environment is done by obstacles present in the warehouse, the proposed approach

13
Intelligent Service Robotics (2021) 14:313–325 319

uses the two essential things into consideration to prevent

the collision. The first thing is to avoid inter clash of robots
with each other. Inter clash of robots is achieved using
thresholding function as defined in 2. There are several
other coordination techniques described by Yan et al. [21].
Another issue to look after is to avoid the obstacle present
in the given workspace. The smart selection technique pro-
posed by the paper handles this issue. An improved smart
selection of pairs to find the nearest best possible destination
for each robot Ri is applied. This Improved smart selection
has an advantage over primitive selection as primitive selec-
tion uses the traditional distance metrics, which does not
consider obstacles. At the same time, the former will find
obstacles. Some more survey on the same issue is presented
[55, 56].

3.3 Target assignment Fig. 5  Target assignment for a goal in the warehouse application

The process of target assignment in a static warehouse envi-

ronment is done by estimating the cost of the path for each traverses the complete given input image between these two
goal from the available robots this cost is obtained by sum- coordinates for finding the best path.
mation of total path length and collisions across the planned
path. The goal is assigned to the robot with minimum cost,
and the process is repeated for all the goals in the system. 4 Results and discussion
A scenario that elucidates target assignment is shown in
Fig. 4, which depicts the warehouse multi-robot environ- The E-Commerce industry is one of the sectors which is
ment. It consists of robots initial position, goal positions, continuously expanding at a very rapid pace, and to cope up
storage pods and depot pods. Figure 5 depicts that p11 is with this requirement, the role of their warehouse becomes
the path from robot R1 to to G1 with no collision and hav- most important. They are deploying robots to meet imme-
ing minimum path. Similarly, p21 is the path from robot R2 diate and accurate demands. KIVA [57, 58] system is one
to G1 with no collision and p31 is the path from robot R3 to of the excellent examples of having an intelligent ware-
G1 with no collision. A distance metric approach is used to house. This paper has utilized the proposed concept of a
smart approach in achieving coordinated path planning in a
warehouse application. Usually, in any intelligent warehouse
application, after each robot receives the corresponding
order task, it needs to plan the path from the current posi-
tion to the target pod and then from the pod position to the
working station. These paths are required to not conflict with

Fig. 4  A pictorial representation of multi-robot system in the ware- Fig. 6  Figure representing the distance ratio of different robots from
house environment a goal

13
320 Intelligent Service Robotics (2021) 14:313–325

the paths of other robots, and there is no deadlock between obtained are displayed in Tables 2, 3, 4, and 5 The execu-
the robots during driving. tion snap is shown in Figs. 7 and 8. The results obtained by
The proposed approach is applied in a static warehouse applying the algorithm on a different multi-robot system by
environment map with a group of 15 robots and 100 stor- varying the number of robots and pods. The results obtained
age pods in the system. In the implementation, the robot is are evaluated using average path length and total computa-
assumed to be a unit sized object and experimental results tion time elapsed as the parameters.

Fig. 7  Simulation sneak peak of warehouse application having 48 pods and 16 robots approaching to reach their respective destinations, green
dots in the picture are the robots moving in the workspace

Fig. 8  Simulation sneak peek of warehouse application having 800 pods and 16 robots moving in the warehouse workspace and approaching to
reach their respective destinations

13
Intelligent Service Robotics (2021) 14:313–325 321

Table 2  Experimental results obtained for multi-robot system in warehouse environment with 1-to-1 Pod using proposed approach
S.no No. of pods No. of robots Source(S) Destination(S) Path Length Elapsed time (Sec) A* Algo (Sec) ILP Algo (Sec)

1 1 1 (36, 36) (45, 45) 18 0.363433 0.4179 0.4070

2 2 2 (17, 40) (29, 11) (10, 54) (14, 44) 21, 36 2.369620 2.7250 2.6539
3 4 4 (27, 13) (11, 49) (8, 8) (17, 23) 24, 32 4.941293 5.6823 5.5341
(25, 46) (19, 6) (8, 2) (15, 15) 61, 13
4 8 8 (21, 22) (29, 16) (5, 50) (18, 44) 44, 39 8.208342 9.4395 9.1932
(27, 30) (25, 36) (5, 3) (24, 27) 49, 10
(7, 35) (20, 19) (10, 41) (22, 35) 09, 18
(13, 16) (9, 58) (4, 47) (10, 2) 42, 57
5 16 16 (7, 6) (3, 46) (20, 30) (8, 29) 37, 22 17.501243 20.1264 19.6013
(13, 25) (29, 53) (6, 50) (24, 47) 32, 11
(21, 43) (29, 43) (20, 48) (9, 39) 06, 24
(22, 61) (19, 13) (2, 27) (24, 53) 64, 45
(4, 46) (25, 6) (21, 15) (23, 50) 48, 46
(19, 48) (12, 22) (2, 50) (5, 39) 19, 24
(5, 40) (25, 12) (17, 18) (5, 44) 34, 52
(12, 37) (3, 52) (17, 54) (5, 17) 22, 37

Table 3  Experimental results obtained for multi-robot system in warehouse environment with doublings pods using proposed approach
S.no No. of pods No. of robots Source(S) Destination(S) PathLength Elapsed time (Sec) A* Algo (Sec) ILP Algo (Sec)

1 2 1 (19, 60) (11, 45) 23 0.298725 0.3435 0.3345

2 4 2 (4, 34,) (26, 43) (9, 56) (15, 18) 27, 36 0.664236 0.7638 0.7439
3 8 4 (28, 28) (8, 16) (14, 21) (5, 2) 21, 17 0.919797 1.0576 1.0300
(2, 13) (26, 38) (9, 21)(21, 42) 15, 09
4 16 8 (1, 39) (8, 40) (17, 3) (17, 11) 52, 38 17.594425 20.2335 19.7057
(10, 31) (7, 50) (4, 11) (10,32) 26, 21
(27, 53) (25, 11) (20, 32) (24, 51) 28, 41
(23, 4) (11, 43) (21, 44) (5, 57) 42, 20
5 32 16 (19, 46) (11, 58) (9, 56) (5, 5) 20, 59 79.338232 91.2389 88.8587
(14, 31) (25, 56) (6, 12) (17, 42) 27, 22
(9, 19) (5, 31) (15, 14) (8, 54) 11, 26
(25, 47) (20, 43) (22, 26) (6, 39) 24, 18
(29, 18) (25, 58) (17, 54) (21, 57) 48, 05
(19, 1) (27, 44) (10, 54) (6, 56) 62, 29
(3, 61) (7, 17) (18, 2) (3, 57) 74, 44
(25, 8) (26, 11) (22, 53) (21, 42) 48, 36

A comparison of the proposed approach with the A* algo- association with the GA (genetic algorithm), and another
rithm when implemented on the warehouse system with 800 one is the ILP (integration linear programming) method.
pods and 16 robots is done. A graphical representation indi- ILP is an integer linear programming method, which appears
cating a comparison of average path length and computa- to be one of the fastest optimal solver’s. Here, the hybrid
tion time is shown in Tables 6 and 7 and in Figs. 9 and 10. combination of A* & GA is considered; GA is used for task
On comparing average path length, the proposed approach allocation, while A* was used for finding the collision-free
shows little better performance of up to 2.5 % while in terms optimal path. ăThe proposed work results better in terms of
of computation time, a useful improvement of around 13 % elapsed time.
is found. The elapsed time factor of the proposed approach In a large-scale warehouse environment, the position of
is compared with two other approaches in Table 2, 3, 4, the pod often changes since the replenishment tasks and the
and 5. The first approach which is being compared is A* in picking tasks are continuously in progress. The pods placed

13
322 Intelligent Service Robotics (2021) 14:313–325

Table 4  Experimental results obtained for multi-robot system in warehouse environment with tripling pod using proposed approach
S.no No. of pods No. of robots Source(S) Destination(S) Path length Elapsed time (Sec) A* Algo (Sec) ILP Algo (Sec)

1 3 1 (27, 39) (24,59) 23 0.545130 0.6268 0.6105

2 6 2 (21, 19) (27,13) (23, 60) (17, 6) 43, 17 3.711996 4.2686 4.1573
3 12 4 (27, 56) (19, 25) (24, 57) (16, 3) 4, 25 11.287250 12.9802 12.6416
(13, 38) (20, 13) (15, 3) (17, 53) 37, 43
4 24 8 (29, 57) (7, 32) (5, 23) (14, 9) 48, 30 17.132150 19.7019 19.1879
(13, 28) (3, 49) (11, 26) (8, 12) 4, 42
(11, 28) (6, 25) (20, 15) (4, 33) 22, 10
(6, 61) (26, 29) (6, 42) (17, 3) 19, 35
5 48 16 (14, 13) (2, 46) (23, 32) (18, 38) 28, 24 57.878143 66.5598 64.8234
(17, 22) (23, 61) (22, 42) (6, 47) 25, 31
(18, 46) (18, 22) (10, 29) (11, 3) 25, 26
(29, 37) (7, 20) (16, 54) (9, 48) 30, 30
(20, 58) (28, 26) (2, 29) (15, 32) 47, 19
(1, 26) (16, 43) (4, 41) (16, 33) 18, 10
(20, 25) (25, 1) (6, 39) (20, 32) 28, 36
(15, 34) (4, 16) (15, 8) (6, 3) 26, 15

Table 5  Experimental results obtained for multi-robot system in warehouse environment with 800 pod using proposed approach
S.no No. of pods No. of robots Source(S) Destination(S) PathLength Elapsed time (Sec) A* Algo (Sec) ILP Algo (Sec)

1 800 1 (20, 49) (15, 5) 49 1.395674 1.6049 1.5630

2 800 2 (10, 43) (28, 59) (21, 5) (22,3) 49, 62 3.421114 3.9342 3.8316
3 800 4 (18, 61) (27, 25) (8, 35) (8, 41) 36, 35 7.951920 9.1446 8.9061
(29, 17) (6, 61) (15, 14) (18, 6) 17, 67
4 800 8 (19, 42) (26, 16) (12, 56) (9, 12) 21, 21 11.1265 12.7954 12.4616
(26, 43) (12, 1) (17, 17) (15, 51) 35, 53
(15, 43) (19, 20) (12, 23) (15, 48) 23, 32
(23, 16) (29, 53) (24, 20) (10, 39) 5, 33
5 800 16 (24,58) (5, 61) (23, 35) (3, 8) 24, 55 75.545151 86.8768 84.6105
(28, 48) (13, 1) (21, 17) (22, 44) 38, 52
(2, 37) (7, 12) (11, 36) (20, 14) 10, 15
(7, 30) (1, 60) (11, 59) (17, 60) 33, 16
(1, 31) (11, 28) (16, 24) (11, 41) 22, 13
(29, 57) (24, 1) (8, 38) (15, 42) 40, 50
(3, 4) (17, 16) (22, 54) (20, 42) 69, 39
(28, 52) (26, 50) (8, 50) (18, 18) 22, 40

Table 6  Multi-robot system using proposed approach in warehouse Table 7  Experimental results obtained for Multi-robot system using
environment with a set of 16 robots and varying the number of pods A* algorithm and proposed approach in warehouse environment with
a set of 16 robots and 800 pods
Warehouse Specification Comp. Time Average Path
Name of Algorithm Comp. Time Average Path
48 pods and 16 robots 37.577 28.32
800 pods and 16 robots 165.545 32.34 A* Algorithm 190.79 33.1875
Proposed Approach 165.545 32.34

13
Intelligent Service Robotics (2021) 14:313–325 323

simultaneously becomes an arduous task to carry-on. Only

coordination can be achieved in a much more straightfor-
ward way by keeping aside the issue of path length. Simi-
larly, by flipping the issue, optimizing the path length of
individual robots can be done if there are no constraints for
robots not to collide with each other or share their sections
of the path with other robots. However, in a multi-robot
system, these two problems are adjacent to each other and
are required to deal parallely to automate any application
efficiently. Each parameter is needed to handle acutely while
optimizing the other. Here, in this paper, we have taken the
warehouse application to solve the problem. The proposed
approach of smart selection is applied to observe and vali-
Fig. 9  Average path length variations graph of four different multi- date the results.
robot systems with varying pods in number The primary objective is to find an optimal, collision-
free path using smart selection for target assignment and
distance metric path planning and coordination. A custom
algorithm for multi-robot path planning in a static warehouse
environment is created, which facilitates completeness and
coordination for the system. The algorithm is further modi-
fied to generate source and destination pairs out of given
configurations based on distance ratio values and generate
path accordingly, maintaining coordination. The proposed
approach shows the improved results experimented in two
different conditions of warehouse application. The ware-
house application having 48 pods along with 16 robots, and
another scenario with 800 pods along with 16 robots are
considered and tested. Computation time and average path
length are observed, which showed better performance and
Fig. 10  Elapsed time variations graph of four different multi-robot is also compared with the A* algorithm and ILP method in
systems with varying pods in number a tabular form in the result section.
Future prospects include path planning and coordination
in a more complex environment. The proposed approach
in the storage area are transported by robots to the replenish- considers the distance ratio using a single dimension.
ment stations or picking stations for corresponding opera- While in the case of a map with a larger obstacle having
tions. Thus, the original storage position of the pod can be width more than length. Then, this approach would fail to
free to travel by other robots. The thresholding function will find the optimal robot target pair. The scenario considered
avoid any obstacle in the path. The proposed distance metric to solve the path planning problem is a static warehouse,
based smart approach of self-identifying the destination will while in real-time, warehouses are seldom static. There’s
help each robot in making the system more efficient. an ever-changing aspect to any regular warehouse, and
Figures 7 and 8 display the picture of the experiment car- this can be reflected in future work. Also, the workspace
ried out for a multi-robot system on warehouse application. is restricted to a limit of 800 pods which is still smaller
The green color symbol in the figure shows the robot mov- than an average warehouse limit in real-time. In future,
ing inside the warehouse, the black square box is showing efforts can be extended to include larger warehouses which
the pod, and the red square box is the destination where the require higher computing and processing power. This leads
robot is destined to reach. to a more significant problem, equipping each robot with a
processor powerful enough to simulate the real-time ware-
house environment and use it for the computation required.
5 Conclusion and future work Thus, moving to a centralized cloud system would be prag-
matic in the coming time and would be the future extension
This paper addressed the issue of path planning with of this work.
coordination in a multi-robot system. Both the problem
is quite critical in its own. Combining these two to solve

13
324 Intelligent Service Robotics (2021) 14:313–325

References 21. Yan Z, Jouandeau N, Cherif AA (2013) A survey and analysis of

multi-robot coordination. Int J Adv Robot Syst 10(12):399
22. Parker LE (2007) Distributed intelligence: overview of the field
1. D’Andrea R (2012) Guest editorial: a revolution in the warehouse:
and its application in multi-robot systems. In: AAAI fall sympo-
a retrospective on kiva systems and the grand challenges ahead.
sium: regarding the intelligence in distributed intelligent systems.
IEEE Trans Autom Sci Eng 9(4):638–639
pp 1–6
2. Ng M-K, Chong Y-W, Ko K-M, Park Y-H, Leau Y-B (2020) Adap-
23. Capitan J, Spaan MT, Merino L, Ollero A (2013) Decentralized
tive path finding algorithm in dynamic environment for warehouse
multi-robot cooperation with auctioned pomdps. Int J Robot Res
robot. Neural Comput Appl 32(17):13155–13171
32(6):650–671
3. Yang Y, Zhang J, Liu Y, Song X (2019) Multi-agv collision
24. Ota J (2006) Multi-agent robot systems as distributed autonomous
avoidance path optimization for unmanned warehouse based on
systems. Adv Eng Inform 20(1):59–70
improved ant colony algorithm. International conference on bio-
25. Nerurkar ED, Zhou KX, Roumeliotis SI (2011) A hybrid estima-
inspired computing: theories and applications. Springer, New
tion framework for cooperative localization under communication
York, pp 527–537
constraints. In: 2011 IEEE/RSJ international conference on intel-
4. Wen S, Zheng W, Zhu J, Li X, Chen S (2011) Elman fuzzy adap-
ligent robots and systems. IEEE, pp 502–509
tive control for obstacle avoidance of mobile robots using hybrid
26. Sartoretti G, Wu Y, Paivine W, Kumar TS, Koenig S, Choset H
force/position incorporation. IEEE Trans Syst Man Cybern Part
(2019) Distributed reinforcement learning for multi-robot decen-
C Appl Rev 42(4):603–608
tralized collective construction. Distributed autonomous robotic
5. Canny J (1988) The complexity of robot motion planning. MIT
systems. Springer, New York, pp 35–49
press, Cambridge
27. Baxter JL, Burke EK, Garibaldi JM, Norman M (2009) Shared
6. Cho D-H, Jang D-S, Choi H-L (2020) Memetic algorithm-based
potential fields and their place in a multi-robot co-ordination tax-
path generation for multiple dubins vehicles performing remote
onomy. Robot Auton Syst 57(10):1048–1055
tasks. Int J Syst Sci 51(4):608–630
28. Gerkey BP, Mataric MJ (2002) Sold!: Auction methods for mul-
7. Bakdi A, Hentout A, Boutami H, Maoudj A, Hachour O, Bouzouia
tirobot coordination. IEEE Trans Robot Autom 18(5):758–768
B (2017) Optimal path planning and execution for mobile robots
29. Chen X, Zhang P, Du G, Li F (2018) Ant colony optimization
using genetic algorithm and adaptive fuzzy-logic control. Robot
based memetic algorithm to solve bi-objective multiple trave-
Auton Syst 89:95–109
ling salesmen problem for multi-robot systems. IEEE Access
8. Song B, Wang Z, Zou L (2017) On global smooth path planning
6:21745–21757
for mobile robots using a novel multimodal delayed pso algorithm.
30. Han SD, Yu J (2019) Effective heuristics for multi-robot path
Cogn Comput 9(1):5–17
planning in warehouse environments. In: 2019 International sym-
9. Wu Q, Lin H, Jin Y, Chen Z, Li S, Chen D (2020) A new fall-
posium on multi-robot and multi-agent systems (MRS). IEEE, pp
back beetle antennae search algorithm for path planning of
10–12
mobile robots with collision-free capability. Soft Comput
31. Al-Jarrah R, Shahzad A, Roth H (2015) Path planning and motion
24(3):2369–2380
coordination for multi-robots system using probabilistic neuro-
10. Song R, Liu Y, Bucknall R (2019) Smoothed a* algorithm for
fuzzy. IFAC-PapersOnLine 48(10):46–51
practical unmanned surface vehicle path planning. Appl Ocean
32. Ni J, Wang K, Cao Q, Khan Z, Fan X (2017) A memetic algorithm
Res 83:9–20
with variable length chromosome for robot path planning under
11. Mac TT, Copot C, Tran DT, DeKeyser R (2016) Heuristic
dynamic environments. Int J Robot Autom 32(4):414–424
approaches in robot path planning: a survey. Robot Auton Syst
33. Alotaibi ETS, Al-Rawi H (2018) A complete multi-robot
86:13–28
path-planning algorithm. Autonom Agents Multi-Agent Syst
12. Choset H (2001) Coverage for robotics-a survey of recent results.
32(5):693–740
Ann Math Artif Intell 31(1–4):113–126
34. Warren CW (1990) Multiple robot path coordination using artifi-
13. Brumitt BL, Stentz A (1996) Dynamic mission planning for mul-
cial potential fields. In: Proceedings., IEEE international confer-
tiple mobile robots. In: Proceedings of IEEE international confer-
ence on robotics and automation. IEEE, pp 500–505
ence on robotics and automation, vol 3. IEEE, pp 2396–2401
35. Svestka P, Overmars MH (1995) Coordinated motion planning
14. Wawerla J, Sukhatme GS, Mataric MJ (2002) Collective construc-
for multiple car-like robots using probabilistic roadmaps. In: Pro-
tion with multiple robots. In: IEEE/RSJ international conference
ceedings of 1995 IEEE international conference on robotics and
on intelligent robots and systems, vol 3. IEEE, pp 2696–2701
automation, vol 2. IEEE, pp 1631–1636
15. Cui R, Guo J, Gao B (2013) Game theory-based negotiation for
36. Švestka P, Overmars MH (1998) Coordinated path planning for
multiple robots task allocation. Robotica 31(6):923–934
multiple robots. Robot Auton Syst 23(3):125–152
16. Roldán JJ, Garcia-Aunon P, Garzón M, DeLeón J, DelCerro J,
37. Burgard W, Moors M, Fox D, Simmons R, Thrun S (2000) Col-
Barrientos A (2016) Heterogeneous multi-robot system for map-
laborative multi-robot exploration. In: Proceedings 2000 ICRA.
ping environmental variables of greenhouses. Sensors 16(7):1018
Millennium conference. IEEE international conference on robotics
17. Luo M, Hou X, Yang J (2019) Multi-robot one-target 3d path
and automation. Symposia proceedings (Cat. No. 00CH37065),
planning based on improved bioinspired neural network. In: 2019
vol.1. IEEE, pp 476–481
16th international computer conference on wavelet active media
38. Bennewitz M, Burgard W, Thrun S (2001) Optimizing schedules
technology and Information Processing. IEEE, pp 410–413
for prioritized path planning of multi-robot systems. In: Proceed-
18. Dias MB, Zlot R, Kalra N, Stentz A (2006) Market-based
ings 2001 ICRA. IEEE international conference on robotics and
multirobot coordination: a survey and analysis. Proc IEEE
automation (Cat. No. 01CH37164), vol 1. IEEE, pp 271–276
94(7):1257–1270
39. Sanchez G, Latombe JC (2002) Using a prm planner to compare
19. Yuan J, Wang H, Lin C, Liu D, Yu D (2019) A novel gru-rnn
centralized and decoupled planning for multi-robot systems. In:
network model for dynamic path planning of mobile robot. IEEE
Proceedings 2002 IEEE international conference on robotics and
Access 7:15140–15151
automation (Cat. No. 02CH37292), vol 2. IEEE, pp 2112–2119
20. Luna R, Bekris K.E (2011) Efficient and complete centralized
40. Kalra N, Ferguson D, Stentz A (2005) Hoplites: a market-based
multi-robot path planning. In: 2011 IEEE/RSJ international con-
framework for planned tight coordination in multirobot teams. In:
ference on intelligent robots and systems. IEEE, pp 3268–3275

13
Intelligent Service Robotics (2021) 14:313–325 325

Proceedings of the 2005 IEEE international conference on robot- 50. Wagner G, Choset H (2015) Subdimensional expansion for mul-
ics and automation. IEEE, pp 1170–1177 tirobot path planning. Artif Intell 219:1–24
41. Wurm K.M, Stachniss C, Burgar W (2008) Coordinated multi- 51. Claes D, Oliehoek F, Baier H, Tuyls K, et al., (2017) Decentralised
robot exploration using a segmentation of the environment. In: online planning for multi-robot warehouse commissioning, In:
2008 IEEE/RSJ international conference on intelligent robots and AAMAS’17: Proceedings of the 16th international conference on
systems. IEEE, pp 1160–1165 autonomous agents and multiagent systems. pp 492–500
42. Hollinger G, Singh S, Djugash J, Kehagias A (2009) Effi- 52. Araki B, Strang J, Pohorecky S, Qiu C, Naegeli T, Rus D (2017)
cient multi-robot search for a moving target. Int J Robot Res Multi-robot path planning for a swarm of robots that can both fly
28(2):201–219 and drive. In: 2017 IEEE international conference on robotics and
43. van Den Berg J, Snoeyink J, Lin MC, Manocha D (2009) Central- automation (ICRA), IEEE, pp 5575–5582
ized path planning for multiple robots: optimal decoupling into 53. Bolu A, Korçak Ö (2019) Path planning for multiple mobile robots
sequential plans. In: Robotics: science and systems, vol 2. pp 2–3 in smart warehouse. In: 2019 7th international conference on con-
44. Hollinger G, Singh S (2010) Multi-robot coordination with peri- trol, mechatronics and automation (ICCMA). IEEE, pp 144–150
odic connectivity. In: 2010 IEEE international conference on 54. Kumar NV, Kumar CS (2018) Development of collision free path
robotics and automation. IEEE, pp 4457–4462 planning algorithm for warehouse mobile robot. Procedia Compu
45. Ding XC, Rahmani AR, Egerstedt M (2010) Multi-uav convoy Sci 133:456–463
protection: an optimal approach to path planning and coordina- 55. Almadhoun R, Taha T, Seneviratne L, Zweiri Y (2019) A survey
tion. IEEE Trans Robot 26(2):256–268 on multi-robot coverage path planning for model reconstruction
46. Wagner G, Choset H (2011) M*: A complete multirobot path and mapping. SN Appl Sci 1(8):847
planning algorithm with performance bounds. In: 2011 IEEE/ 56. Cortés J, Egerstedt M (2017) Coordinated control of multi-robot
RSJ international conference on intelligent robots and systems. systems: a survey. SICE J Control Meas Syst Integr 10(6):495–503
IEEE, pp 3260–3267 57. Wurman PR, D’Andrea R, Mountz M (2008) Coordinating hun-
47. Otte M, Correll N (2014) Any-com multi-robot path-planning dreds of cooperative, autonomous vehicles in warehouses. AI Mag
with dynamic teams: multi-robot coordination under communi- 29(1):9–9
cation constraints. Experimental robotics. Springer, New York, 58. Bogue R (2016) Growth in e-commerce boosts innovation in the
pp 743–757 warehouse robot market. Ind Robot Int J
48. Contreras-Cruz MA, Ayala-Ramirez V, Hernandez-Belmonte UH
(2015) Mobile robot path planning using artificial bee colony and Publisher’s Note Springer Nature remains neutral with regard to
evolutionary programming. Appl Soft Comput 30:319–328 jurisdictional claims in published maps and institutional affiliations.
49. Ma H, Wu X, Gong Y, Cui Y, Song J (2015) A task-grouped
approach for the multi-robot task allocation of warehouse sys-
tem. In: 2015 international conference on computer science and
mechanical automation (CSMA). IEEE, pp 277–280

13