Development and Implementation of A Multi-Robot System For Collaborative Exploration and Complete Coverage

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

2018 14th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS)

Development and Implementation of a Multi-Robot System for


Collaborative Exploration and Complete Coverage

Yi-Chun Huang1 and Huei-Yung Lin2


1 Department of Electrical Engineering, Advanced Institute of Manufacturing with High-Tech Innovation, National Chung
Cheng University, Chiayi 621, Taiwan
2 Department of Electrical Engineering, National Chung Cheng University, Chiayi 621, Taiwan

[email protected], [email protected]

Keywords: Multi-robots, Complete Coverage, Path Planning, Cost Function

Abstract: Multi-robot research has significant progress in many aspects. Recently, it tends to reduce the time task type
such as disaster relief and exploration. Multi-robots in large environments and on the factory side are mainly
planned for dispatch tasks. This paper proposes two cost functions and a multi-robot complete coverage
algorithm. This provides a path to keep a robot moving forward at a target point with more area gain. It
enables the robots to completely cover an unknown large-scale environment in a short period of time, and
effectively reduce the over-repetition path that has been explored by other robots. Our approach is validated
using a simulation platform with experimental results performed under complex environments.

1 INTRODUCTION the robot’s two-dimensional grid map appear as a sim-


ulation platform. Through the path optimization and
With the boom in industry and the advancement of excessive repetitive path reduction, the overall effi-
smart appliances, robotics research topics are all- ciency is improved.
encompassing. The complete coverage and path plan- Complete coverage and path planning (CCPP) is
ning are no longer only to meet the past service con- a study that considers coverage and path planning.
ditions, but to expand into the field of medicine, so- It focuses on coverage efficiency, total time to com-
cial security, academic research, agriculture and other plete coverage, total path length and energy consump-
fields which can bring about a completely different tion (Lee et al., 2011; Galceran and Carreras, 2013),
development from the past. In the field of robotics, and is suitable for agriculture, cleaning, monitoring,
multi-robot research is more challenging than deal- etc. The approach is divided into offline and online
ing with only a single robot. People design different CCPP algorithms. The offline algorithm is mainly
conditions according to the type of tasks that can be used for random coverage path planning (Khan et al.,
completed, and use multi-robot cooperation and infor- 2017). Liu et al. (Liu et al., 2008) proposed an online
mation sharing to improve the efficiency of the robots. random path planning algorithm, which can be flex-
The multi-robot systems will then be able to practice ibly applied to smaller unknown environments. The
disaster relief operations, assist staffing, and serve hu- Boustrophedon trapezoidal cell decomposition tech-
mans in reality. nique (Acar et al., 2002) involves the offline cell de-
In this paper, we focus on time minimization for composition algorithm of robots in a polygonal space
complete coverage, reduced coverage, and optimal (Zhou et al., 2012; Kim et al., 2014). The repre-
path planning. It aims to reduce the phenomenon sentation method is based on a unified mesh struc-
of excessive labor, that is, to assist in the search for ture to express the environment type to complete the
“scattered items” or “regularly check the environment local coverage task. This method uses the Voronoi
for abnormalities” in large-scale environments (such diagram(Palacios-Gass et al., 2016), neural network
as airports, opera houses, factories, etc.). Technically, (Shivashankar et al., 2011; Lau et al., 2013) for com-
a group of robots will perform the complete coverage plete coverage and path planning. Considering more
and path planning for an unknown environment, and efficient coverage methods such as minimizing the
a more regular cost function, target selection function number of slopes, time minimization and backtrack-
and related technical strategy should be designed. Let ing mechanisms, A* algorithm (Hart et al., 1968;

978-1-5386-9385-8/18/$31.00 ©2018 IEEE 472


DOI 10.1109/SITIS.2018.00078
Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
Chand and Carnegie, 2012), D* algorithm(Stentz, tive multi-area coverage (CMAC), and let the multi-
1994; Dakulovi et al., 2011) and Theta* algorithm robot complete a complex mission. They use multi-
(Nash et al., 2007; Viet et al., 2013) are available in point dynamic aggregation (MPDA) and collaborative
the trajectory optimization. motion to finish the complete coverage in unknown
Planning the best path for complete coverage in an environments. Some recent works use machine learn-
unknown environment is one of the hot topics in the ing techniques to train the robot and make indepen-
robotics field. Many researchers have studied the de- dent judgments and navigation plans. Pfeiffer et al.
velopment for single robots, and applied the efficient (Pfeiffer et al., 2017) design a model and the robot
algorithms to the multi-robot cooperation tasks with learns the environment information using simulation
some modifications. The work by Burgard et al. (Bur- and synthetic data. It is then applied to the unknown
gard et al., 2002) provides a way to reach the same environment and let the robot plan the path and navi-
target by multi-robot collaborative exploration in un- gate autonomously.
known environments. The basic idea is to navigate In this paper, we focus on minimizing the com-
along the shortest cost path to shorten the exploration putation time for complete coverage and reducing the
time and improve the overall efficiency. They adopt a repeated coverage area by optimal path planning. The
greedy algorithm and add the goal selection and cost objective is to decrease the excessive labor in some
function design. This two concept later becomes an applications such as searching for scattered items or
important reference for a method of efficient explo- regularly check the environment for abnormalities in
ration. large-scale environments (e.g., airports, opera houses,
Senthilkumar and Bharadwaj (Senthilkumar and factories, etc.). Our proposed technique is for com-
Bharadwaj, 2012) provides S-MSTC (ES-MSTC) ac- plete coverage and path planning with a team of
cording to the cases of emergency situations such as robots in an unknown environment. We design a regu-
terrain investigation, search and rescue missions, and lar cost function, a goal selection function and related
intrusion detection. They propose a technique to com- algorithms, and let the robot display on a simulation
pletely cover the unknown environment as soon as platform with a 2D grid map. The overall efficiency
possible, and consider the communication in a robot is optimized through path optimization and duplicate
team to minimize damage rate. Their approach adopts path reduction.
ant crawling properties which leave traces in the cov-
ered terrain. Multiple spanning trees can be created
at the same time, and complete coverage can be per- 2 COST FUNCTION
formed in small areas and narrow door openings.
Viet et al. (Viet et al., 2013) deal with the prob- To design a cost function with less coverage time for a
lem of complete coverage in unknown surroundings. multi-robot system, we consider the factors of sports
They present a novel method that each robot uses mode and best algorithm solution. This can produce a
Boustrophedon motion to cover the unvisited area un- better moving cost function and a target selection cost
til the robot termination position is reached (around function design. We will design a guiding trajectory
the overlaid regions or obstacles). When the robot different from the sweeping robot to perform the max-
reaches its end position, the backtracking point is imum area complete coverage, and practice the full
checked by detecting the past records based on multi- coverage after using the cost function on the simula-
target Theta* algorithm, and plan the shortest trace- tion tool to improve the efficiency. Through the for-
back path to the next starting position. On the other mula in Eq. (1), the robot is able to find the most suit-
hand, Pereira et al. (Pereira et al., 2015) investigate able target range. It is mainly designed to prevent the
the multi-robot distributed exploration to completely trajectory too close to the environment boundary or
cover the unknown environment. The exploration along the wall, which tend to cover too many known
method is based on individual decision rules and the regions and waste the sensing range (typically from
communication via the topology map to minimize laser range scanning).
the spatial overlap. Their approach adopts the well-
known frontier-based exploration (Yamauchi, 1998) θ θ
algorithm, and emphasizes the multi-robot coordina- C(t + 1) = f · (d + ) + f¯ · (1)
tion communication structure, and intelligently se- related f in α(i)
lecting the next target after communication. In the above equation, d is the midpoint of the
In recent years, more and more types of multi- distance between the two obstacles seen in the visi-
robot applications are available and widespread. Xin ble range of the robot, θ is the steering angle of the
et al. (Xin et al., 2017) propose the idea of coopera- robot between -90 and 90 degrees. related f in is the

473

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
information about the obstacles contained in the over- 3 COMPLETE COVERAGE AND
all coverage of the robot. In the case where the cur- PATH PLANNING
rent robot is facing, there is more information on the
obstacle points and the degree of association is higher. In addition to the cost function, we also propose a
All of the above must meet the condition that f equals multi-robot complete coverage algorithm to complete
1 and will become the reference for robot movement. the entire coverage. The main purpose is to reduce the
f is designed to have an index with two states 0 or excessive steering of the robot and return to the initial
1. When robotHead is between −15 and 15 degrees position planning to the next area for unknown cov-
where the obstacle area exceeds the coverage free area erage. For cooperative exploration, each robot broad-
or multiple robots cover the same area that f = 1. casts its location and map to others, and integrates the
f = 0 holds when robotHead between −15 to 15 de- information received from others to create a global
grees, where the coverage free are is more than the map. The global map is updated consistently for com-
obstacle area or the rotate angle over 45 degrees. f¯ is plete multi-robot exploration. The basic multi-robot
the opposite of f . algorithm is shown in Algorithm 2.

Algorithm 2 Basic Multi-Robot Algorithm


Algorithm 1 Goal Algorithm Require: A team of robots
Require: The robot position Ensure: Explored map
Ensure: The robot position 1: repeat
1: if f ==0, then 2: To update map, read visible information from
2: collecting sensor information from -90 to 90, to sensor.
find the place of the most of coverage f ree. 3: Front obstacle detection
3: Set a node every five degree. 4: detecting obstalce,then doing rotation to keep
4: combine 2 and 3 information to find which posi- the biggest distance and following wall to ex-
tion is the best with the most α(i). plored map.
5: moving to the target. 5: Send information to others.
6: if known place to the unkown one, then 6: Receive others explored information.
7: Using Theta* algorithm (Viet et al., 2013) or 7: Updating map
BM tracking algorithm (Viet et al., 2013) to the 8: until All information {obstacle, coverage free} is
bounding of target. filled.
8: else
9: According to cost function to find the matched In order to make the simulation easier to verify
information to coverage completely. the algorithm, we assume that multi-robot communi-
10: else cation is idealized. Thus, it will not be affected by
11: go to 9 Step 5 and Step 6 in Algorithm 2. We then add the
multi-robot cost function sorting, and let the whole
system the most efficient arrangement for each robot.
The design details are shown in Algorithm 3.
If f is zero, it would check what is the best way The robot checks the information and updates the
moving to the next target position, and use α(i) and global map through robotHead in Algorithm 3. It
related f in in this case as shown in Algorithm 1. The contains the important information providing which
simplified goal selection function is given by robot is best suited to cover a specific unknown area.
Considering how to achieve optimal movement of a
multi-robot system, we sort all cost functions and
G(x) = max(α(i)&related f in) (2) cover unknownarea using the one with the smallest
value as the first priority. Other robots recalculate
the cost function and use α(i) to move to the nearest
In Algorithm 1, the information in -90 to 90 degrees unknownarea.
enables the robot to find the next candidate target Our objective is to use multiple robots to com-
point. Checking which direction of the entire cover- pletely cover an unknown large-scale environment. In
age area contains less information (obstacle, no cover- Algorithm 3, there are still cases of coverage failures:
age), and higher related f in and α(i) coverage areas. too dense obstacles and fragmented unknown areas.
We use Theta* algorithm (Viet et al., 2013) to make Therefore, we re-adjust Algorithm 4 by adding three
the robot move to an unknown area. features L1 , L2 , L3 to help the robot find the best target

474

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
Algorithm 3 Multi-Robots Strategy Algorithm
1: if alldata in coverage area is coverage f ree, then
2: check robotHead and explored map
3: else
4: if unknowarea obtain coverage f ree and
obstacledata , then
5: Sort each robot cost function, with high
related f in and low rotation will get the higher
priority.
6: else
7: the robot which move the lowest distance will ex-
plore the area.
8: else
9: Explored the largest α(i) by related f in.
10: if two robot repeated the same area of coverage,
then
11: go to 4
12: else
13: go to 1

point and plan the best path. When the lowest cost is
found, other robots do not choose to cover the same
path in the same area. We want to cover the largest
α(i) unknown area by checking L1 , L2 and L3 , with
the first priority L1 .
Figure 1: Flowchart of the proposed system.
Algorithm 4 Multi-Robots proposed Algorithm
Require: A team of robots
is designed to verify our algorithmic efficiency and
Ensure: Explored map
display the results for the unknown environments.
1: repeat
2: Initialize M=φ
3: Moving by Cost(), updating M ,and label as
L3 if there are not filled information when mov- 4 EXPERIMENTAL RESULTS
ing.
4: Checking robotHead side with the informa- Our experiment contains three main parts. The first
tion during two obstacle having coverage f ree is the simulation, which includes the interface, finite
4.1 Yes, rotating the highest in related f in and conditions, and robotHead with laser scanning. The
label the rotated target as L1 ,the other side label second part shows that our approach works well be-
as L2 . Completing explored L1 will explore the tween a single robot and multiple robots. The last
closet unknown area used BM tracking algorithm one is to use our approach to compare with the co-
or Theta* algorithm. 4.2 No, go to 3. ordination for multi-robot exploration using topolog-
5: checking robots repeated the same area of ical maps (Pereira et al., 2015). The results show that
coverage our method can be used for complete coverage of un-
6: Checking M filled all information {obstacle, known large-scale environments.
coverage free} is filled
We use FLTK for the simulation interface,
7: until All information {obstacle, coverage free} is
OpenGL as the drawing tool, and OpenCV to read the
filled.
binary image. The platform is Visual Studio 2010.
To accommodate the large-scale map style, we design
The system flowchart of our complete coverage the map interface to be 1300 × 900 as shown in Fig-
approach is show on Figure 1. The map is represented ure 2. The binary image is displayed in Figure 2(a).
with binary cells, and proceeded with all steps to fin- Figures 2(b) and 2(c) illustrate the known map and the
ish the complete coverage. The simulation interface complete coverage result.

475

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
(a) Binary map (b) Unknown map (c) Coverage map
Figure 2: The simulation interface.

(a) Work-a (b) Work-b

(a) multi-room (b) classroom

(c) Work-c (d) Work-d


Figure 4: Work style environments.
Table 1: Efficiently of multi robot proposed algorithm
(c) airport-style (d) opera-style
Figure 3: Large-scale environments. Efficiency of proposed algorithm
Single Robot Multi-Robot
Next we introduce our robotHead laser scan. classroom 128.524 seconds 64.837 seconds
Usually, it has four directions (front, left, right, rear). airport-style 370.352 seconds 184.496 seconds
In our case, we also consider the front left, front right, Work-b 406.691 seconds 202.382 seconds
rear left and rear right eight directions. How to choose Work-a 263.367 seconds 115.959 seconds
the direction is to check robotHead according to the
cost function, and get the 0 - 180 degree data of the
objective function as the next candidate choice. The based algorithm combined with a topology map for
designed map is shown in Figures 3 and 4. full coverage of multiple robots. The results of both
We used four images, Figures 3(b), 3(c), 4(a) and methods are shown in Table 2. Figure 9 shows the
4(b), to evaluate the efficiency of our algorithm un- results of complete coverage. Our approach can
der a single robot and multiple robots, and the results reduce the total processing time and have better path
are shown in Table1. The complete coverage simula- planning.
tion results for a single robot and multiple robots are
shown in Figure 5. Figure 6 is a complete coverage
of the unknown environment with the maximum α(i)
that both robots remain fully covered. Figures 7 and 8 5 CONCLUSIONS
show the complete coverage of multiple robots in an
unknown large environment and complex workplace. This paper proposes a new cost function and multi-
The trajectory of the robot is clear and the overlap robot complete coverage strategy design, which can
with other robots is reduced. completely cover the unknown and complex factory
We compare our method with coordination environment on the simulation side. There are sev-
for multi-robot exploration using topological eral contribution in the research: 1) The motion cost
maps(Pereira et al., 2015). It uses a leading edge function is proposed to make the robot move with-

476

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
REFERENCES
Acar, E. U., Choset, H., Rizzi, A. A., Atkar, P. N., and
Hull, D. (2002). Morse decompositions for cover-
age tasks. The International Journal of Robotics Re-
search, 21(4):331–344.
Burgard, W., Moors, M., and Schneider, F. (2002). Col-
laborative exploration of unknown environments with
(a) single robot classroom (b) multi-robot classroom teams of mobile robots. In Beetz, M., Hertzberg, J.,
Ghallab, M., and Pollack, M. E., editors, Advances in
Plan-Based Control of Robotic Agents, pages 52–70,
Berlin, Heidelberg. Springer Berlin Heidelberg.
Chand, P. and Carnegie, D. A. (2012). A two-tiered global
path planning strategy for limited memory mobile
robots. Robotics and Autonomous Systems, 60(2):309
– 321.
(c) single robot airport-style (d) multi-robot airport-style Dakulovi, M., Horvati, S., and Petrovi, I. (2011). Complete
coverage d* algorithm for path planning of a floor-
cleaning mobile robot. IFAC Proceedings Volumes,
44(1):5950 – 5955. 18th IFAC World Congress.
Galceran, E. and Carreras, M. (2013). A survey on coverage
path planning for robotics. Robotics and Autonomous
Systems, 61(12):1258 – 1276.
Hart, P. E., Nilsson, N. J., and Raphael, B. (1968). A for-
(e) single robot Work-b (f) multi-robot Work-b mal basis for the heuristic determination of minimum
cost paths. IEEE Transactions on Systems Science and
Cybernetics, 4(2):100–107.
Khan, A., Noreen, I., and Habib, Z. (2017). On complete
coverage path planning algorithms for non-holonomic
mobile robots: Survey and challenges. J. Inf. Sci.
Eng., 33(1):101–121.
(g) single robot Work-a (h) multi-robot Work-a Kim, D. H., Hoang, G., Bae, M., Kim, J. W., Yoon, S. M.,
Yeo, T., Sup, H., and Kim, S. (2014). Path tracking
Figure 5: Completed coverage on our interface.
control coverage of a mining robot based on exhaus-
tive path planning with exact cell decomposition. In
2014 14th International Conference on Control, Au-
tomation and Systems (ICCAS 2014), pages 730–735.
out relying on the traditional motion mode. 2) The
Lau, B., Sprunk, C., and Burgard, W. (2013). Efficient
goal selection cost function is proposed, which allows
grid-based spatial representations for robot naviga-
the robot to select the best favorable conditions as the tion in dynamic environments. Robotics and Au-
moving target. 3) An easy-to-use simulation interface tonomous Systems, 61(10):1116 – 1130. Selected Pa-
is designed to let users verify the efficiency of the al- pers from the 5th European Conference on Mobile
gorithm. Robots (ECMR 2011).
Lee, T.-K., Baek, S.-H., Choi, Y.-H., and Oh, S.-Y. (2011).
Smooth coverage path planning and control of mobile
robots based on high-resolution grid map representa-
tion. Robotics and Autonomous Systems, 59(10):801
– 812.
ACKNOWLEDGEMENTS
Liu, Y., Lin, X., and Zhu, S. (2008). Combined coverage
path planning for autonomous cleaning robots in un-
structured environments. In 2008 7th World Congress
The support of this work in part by the Ministry on Intelligent Control and Automation, pages 8271–
8276.
of Science and Technology of Taiwan under Grant
Nash, A., Daniel, K., Koenig, S., and Feiner, A. (2007).
MOST 106-2221-E-194-004 and the Advanced In-
Theta*: Any-angle path planning on grids. In Pro-
stitute of Manufacturing with High-tech Innovations ceedings of the 22Nd National Conference on Artifi-
(AIM-HI) from The Featured Areas Research Cen- cial Intelligence - Volume 2, AAAI’07, pages 1177–
ter Program within the framework of the Higher Ed- 1183. AAAI Press.
ucation Sprout Project by the Ministry of Education Palacios-Gass, J. M., Montijano, E., Sags, C., and Llorente,
(MOE) in Taiwan is gratefully acknowledged. S. (2016). Distributed coverage estimation and control

477

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
(a) T=0 (b) T=20 (c) T=50

(d) T=75 (e) T=90 (f) T=128


Figure 6: The result of large-scale unknown environment with complete coverage.
Table 2: The result of compared with two algorithm.

Compared with two algorithm


Method of (Pereira et al., 2015) My method Decrease time%
Image A 187.252 seconds 118.347 seconds 37.8%
Image B 168.816 seconds 110.924 seconds 34.3%
Image C 203.577 seconds 184.496 seconds 9.4%
Image D 172.378 seconds 140.941 seconds 19.2%

(a) multi-robot Work-a (b) multi-robot Work-b


(a) multi-robot multi-room (b) multi-robot classroom

(c) multi-robot Work-c (d) multi-robot Work-d


(c) multi-robot airport-style (d) multi-robot opera-style Figure 8: Completed coverage on our interface.
Figure 7: Completed coverage on large-scale environment.

data-driven approach to end-to-end motion planning


for multirobot persistent tasks. IEEE Transactions on for autonomous ground robots. In 2017 IEEE In-
Robotics, 32(6):1444–1460. ternational Conference on Robotics and Automation
(ICRA), pages 1527–1533.
Pereira, T., Moreira, A. P., and Veloso, M. (2015). Coor-
dination for multi-robot exploration using topological Senthilkumar, K. and Bharadwaj, K. (2012). Multi-
maps. In Moreira, A. P., Matos, A., and Veiga, G., robot exploration and terrain coverage in an unknown
editors, CONTROLO’2014 – Proceedings of the 11th environment. Robotics and Autonomous Systems,
Portuguese Conference on Automatic Control, pages 60(1):123 – 132.
515–524, Cham. Springer International Publishing. Shivashankar, V., Jain, R., Kuter, U., and Nau, D. S. (2011).
Pfeiffer, M., Schaeuble, M., Nieto, J., Siegwart, R., and Real-time planning for covering an initially-unknown
Cadena, C. (2017). From perception to decision: A spatial environment. In Murray, R. C. and McCarthy,

478

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.
tional Conference on Autonomous Agents, AGENTS
’98, pages 47–53, New York, NY, USA. ACM.
Zhou, P., Wang, Z.-m., Li, Z.-n., and Li, Y. (2012). Com-
plete coverage path planning of mobile robot based
on dynamic programming algorithm. In 2nd interna-
tional conference on electronic and mechanical en-
gineering and information technology, pages 1837–
1841.

(a) Image A (Pereira et al., (b) Image A


2015)

(c) Image B (Pereira et al., (d) Image B


2015)

(e) Image C (Pereira et al., (f) Image C


2015)

(g) Image D (Pereira et al., (h) Image D


2015)
Figure 9: Comparison of two algorithms with four different
maps.

P. M., editors, Proceedings of the Twenty-Fourth In-


ternational Florida Artificial Intelligence Research
Society Conference, May 18-20, 2011, Palm Beach,
Florida, USA. AAAI Press.
Stentz, A. (1994). Optimal and efficient path planning for
partially-known environments. In Proceedings of the
1994 IEEE International Conference on Robotics and
Automation, pages 3310–3317 vol.4.
Viet, H. H., Choi, S., and Chung, T. (2013). An online
complete coverage approach for a team of robots in
unknown environments. In 2013 13th International
Conference on Control, Automation and Systems (IC-
CAS 2013), pages 929–934.
Xin, B., Gao, G., Ding, Y., Zhu, Y., and Fang, H. (2017).
Distributed multi-robot motion planning for coopera-
tive multi-area coverage. In 2017 13th IEEE Inter-
national Conference on Control Automation (ICCA),
pages 361–366.
Yamauchi, B. (1998). Frontier-based exploration using mul-
tiple robots. In Proceedings of the Second Interna-

479

Authorized licensed use limited to: VIT University- Chennai Campus. Downloaded on December 21,2023 at 14:43:03 UTC from IEEE Xplore. Restrictions apply.

You might also like