See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329033819

Simatch: A Simulation System for Highly Dynamic Confrontations Between

Multi-Robot Systems

Conference Paper · November 2018

DOI: 10.1109/CAC.2018.8623698

CITATIONS READS
4 690

6 authors, including:

Zhiqian Zhou Junchong Ma

16 PUBLICATIONS 66 CITATIONS
National University of Defense Technology
4 PUBLICATIONS 101 CITATIONS
SEE PROFILE
SEE PROFILE

Huimin Lu Junhao Xiao

National University of Defense Technology National University of Defense Technology
129 PUBLICATIONS 1,201 CITATIONS 20 PUBLICATIONS 162 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Huimin Lu on 19 November 2018.

The user has requested enhancement of the downloaded file.

 Simatch: A Simulation System for Highly Dynamic
Confrontations between Multi-Robot Systems
Zhiqian Zhou,Weijia Yao, Junchong Ma, Huimin Lu, Junhao Xiao, Zhiqiang Zheng
Department of Automation
National University of Defense Technology
Changsha, P. R. China
setaria_viridis@126.com, weijia.yao.nudt@gmail.com, junchong_nubot@outlook.com,
lhmnew@nudt.edu.cn, junhao.xiao@ieee.org, zqzheng@nudt.edu.cn

Abstract—Simulation is becoming more and more important another simulator in RoboCup Simulation, focus on rigid
for robotics research, especially for multi-robot system research. body dynamics and the simulation of a variety of sensors
Simatch, originating from the Robot World Cup (RoboCup) and actuators [3]. However, it is not widely used in
Middle Size League (MSL) match, is proposed to simulate
the highly dynamic confrontation between multi-robot systems other domains. USARSim (Unified System for Automation
and validate the adversarial multi-robot strategies. It consists and Robotics Simulation), combined with ROS (Robot
of three independent subsystems: the simulation subsystem Operating System), has been used in various applications
setting up the simulated environment, the strategy subsystem since 2012 [4]. But it is only commercially available.
realizing the multi-robot strategies and the scene subsystem In [5], a simulation system for MSL robots has been
describing the specific scene. Due to the independence of
these subsystems, it is easy to employ Simatch for various realized based on Gazebo, but the project has been out
applications. Simatch is validated by a series of tests, including of maintenance.
a test of the omnidirectional locomotion, a simulated MSL Considering the fast implementation and perfect inter-
match among ten robots and a simulated encirclement between face with real robot codes based on ROS, Simatch, a
sixteen robots. Since 2016, Simatch has been the official brand new simulation system based on Gazebo and ROS,
software for the MSL simulation match in the Chinese Robotics
Competition. Besides, an MSL simulation project based on is built to simulate an MSL match, or any other kinds of
Simatch has been proposed to promote the development of confrontation between multi-robot systems. This paper is
MSL in the 2017 MSL international workshop. The proposed an important extension of [6] and [7], which only introduce
simulation system facilitates the development of multi-robot the implementation of the simulation environment and
cooperation algorithms and related research. the realization of basic motions of simulation models, etc.
Index Terms—Simatch, simulation system, highly dynamic,
adversarial, multi-robot system, strategy. In this paper, an extended simulation system consisting
of several new components is developed to facilitate the
I. Introduction research of multi-robot systems. It is notable that Simatch
Middle Size League (MSL) [1] is one of the founding comes with many new features. It can not only be used to
soccer leagues of Robot World Cup (RoboCup)1 . In this simulate the confrontation between multi-robot systems
league, two teams each comprising five autonomous robots but also record all the data of the game as a data set. In
play against each other in a soccer match according to the addition, the game can be completely reproduced using
modified game rules based on the actual human soccer the data-playback function. Moreover, Simatch is open
game. The novel feature of MSL is its highly dynamic source and highly flexible and extensible, it has been used
and aggressive environment. Robot players with average for research on encirclement control [8], [9], multi-robot
weight nearly 40 kg can run at a speed over 5 m/s. It cooperation [10], task allocation [11] and so on.
is a high-level testbed for multi-robot strategies. Since This paper is organized as follows. In Section II, the
collision is common in this game and robots are quite overall architecture of Simacth is depicted. Section III-V
complex and expensive, experiments on real robots are describe the structure and function of three subsystems.
very costly and difficult. Therefore, we turn to build a Then, Section VI introduces two key interfaces combining
simulation system to simulate the highly dynamic and these independent subsystems. Various validations are
aggressive match between multi-robot systems. presented in Section VII. Section VIII concludes the paper
There are some available robot simulation systems. and summarizes the future work.
Übersim [2] simulates the dynamic and aggressive en- II. Overall Architecture
vironment in RoboCup Small Size League (SSL). The
main deficiency of Übersim is that the robot models Fig. 1 shows the general architecture of Simatch.
can only be parameterized at compiling time. SimRobot, Simatch includes two teams of robots, “Cyan Team” and
“Magenta Team”, and three independent subsystems, the
1 http://http://www.robocup.org/ simulation subsystem, the strategy subsystem and the
scene subsystem. These subsystems are highly indepen- in Table 1. The robot model and the simulation world are
dent. A brief introduction for each subsystem is described shown in Fig. 2.
in the following text. TABLE I: Parameters for a simulation MSL match
P roperty N ame V alue
physics engine Open Dynamics Engine (ODE)
max step size 0.005
update rate 200
gravity 9.8
ground_plane, soccer field,
model plugins
left goal, right goal

Fig. 1: The general architecture of Simatch.

The simulation subsystem is the basis of Simatch. It
models the simulated environment with various simulation
models based on Gazebo and realizes basic motions of
simulation models. By modifying simulation models, it is
possible to simulate various robots and environments. It
(a) (b)
also provides exact statuses of models for other subsys-
tems, which is the basis of a multi-robot collaboration.
The strategy subsystem realizes multi-robot strategies.
Generally speaking, the strategy subsystem does not
depend on a specific robot model or multi-robot system,
which makes it possible to test multi-robot strategies for
real robots on a simulated multi-robot system. Each team
of robots have their own strategies. By sending velocity (c)
commands, it controls simulated robots’ motions. “World Fig. 2: Mesh property (a) and collision property (b) of the
Model” is a virtual world built by a team of robots, robot model and the simulation world (c).
including statuses of simulation models and strategies B. Realization of the Omnidirectional Locomotion
information of the team. Compared with the traditional nonholonomic dual-drive
The scene subsystem is built to describe the specific wheeled robot, the omnidirectional mobile robot is able
scene and control the process of a game with game to synchronize steering and linear motion in any direction
commands that divide a complicated or simple game into [12]. This advantage improves the flexibility of the robot
a series of stages. They help two teams of robots to agree greatly and enables the robot to realize faster target track-
on the process of the game, which is the key to realize the ing and obstacle avoidance, which is extraordinarily im-
confrontation. The game command is built to describe portant in a highly dynamic and aggressive environment.
different stages and it is also the only shared information Therefore, nearly all teams employ omnidirectional motion
for two teams of robots. system in real MSL matches. To improve the flexibility of
III. Simulation Subsystem simulated robots and enhance the antagonistic between
The simulation subsystem has two important functions. simulated robots, the omnidirectional motion based on
The first one is to set up a simulated environment, which our real robot [13] is simulated.
consists of various simulation models and a simulation
world. Another one is to realize basic motions of simulation
models.
A. Simulation Models and the Simulation World
In [6], a typical environment for MSL matches is built. (a) (b)
Its simulation models include the robot model, the soccer
field model, the goal model and the soccer ball model. Fig. 3: Omnidirectional wheel and motion model of the
They are spawned as model plugins of Gazebo. The wheel (a) and base frame of the real robot (b).
realization of basic motions, including omnidirectional Our custom-designed omnidirectional wheel and its
locomotion, ball-dribbling and ball-kicking, is written in motion model are shown in Fig. 3(a) and the base frame
the corresponding model plugin. The simulation world with four omnidirectional wheels is illustrated in Fig. 3(b).
determines lighting, simulation step size, simulation fre- The omnidirectional wheel consists of a motor-controlled
quency and other simulation properties, which are shown wheel hub and sixteen passive rollers. Rollers rotate in
the vertical direction perpendicular to the hub rotation TABLE II: Parameters for the omnidirectional motion
axis, which enables the wheel to move smoothly in any symbol α1 α2 α3 α4 β d
direction. The radius of omnidirectional wheels is denoted value π 3π
− 3π − π4 π
0.203(m)
4 4 4 2
by r and the rotation speed of the ith wheel hub is
represented by ωi .
IV. Strategy Subsystem
The strategy subsystem is built to realize multi-robot
strategies, including multi-robot collaboration, multi-
robot path planning, task allocation and so on. Consid-
ering the interface with real robot codes, the framework
of the strategy subsystem is built as a distributed multi-
robot strategy system based on ROS. Its framework is
shown in Fig. 5.
(a) (b)
Fig. 4: Layout of omnidirectional locomotion system (a)
and coordinate frames (b).
Fig. 4(a) shows the layout of our omnidirectional loco-
motion system in two dimension plane. The gray rectan-
gles represent four omnidirectional wheels with numbers
from 1 to 4. di denotes the distance between the center
of the ith wheel and the center of the robot. The robot
coordinate frame is presented by xoy with its origin at the
center of the robot.
Fig. 4(b) shows coordinate frames of each omnidirec-
tional wheel. xoy is the robot coordinate frame and x′i o′i yi′
is the ith wheel coordinate frame attached to the center of
the ith ominidirectional wheel. o′i x′i is the rotation center Fig. 5: The framework of the strategy subsystem.
axis of motor hub and o′i yi′′ is parallel to the rotation center
axis of passive rollers and o′i x′′i is perpendicular to o′i yi′′ . αi The strategy subsystem is composed of several types
is the angle between ox and oo′i . β i is the deflection angle of nodes. The cyan rectangles stand for nodes and the
of the roller, representing the angle from o′i x′i to o′i yi′′ . γ i arrows represent information flow based on topics. Each
is the rotation angle between xoy and x′i o′i yi′ . node is a process for a specific task and all nodes
The speed of the robot is described by (vx , vx , ω)T in are combined together into a graph. Its communication
xoy. We define W = (w1 , w2 , w3 , w4 )T to describe rotation infrastructure mainly depends on the ROS middleware,
speeds of four wheels. including publish/subscribe anonymous message passing
Since four omnidirectional wheels share the same struc- and request/response remote procedure calls2 . The main
ture, βi shares the same value, which is denoted by β. All nodes are:
di are decided by the layout of omnidirectional motion Gazebo : The Gazebo node is created by the simulation
system and they share the same value d. Besides, o′i x′i subsystem. With the package named gazebo_ros_pkgs3 ,
coincides with oo′i in our real robots, then: the simulation subsystem publishes topic ”OmniVision-
Info” and provides exact statuses of simulation models
γi = αi , i = 1, 2, 3, 4. (1)
for other subsystems.
According to the kinematic analysis of the omnidirec- W orldM odel : The Nubot Control node, the World
tional locomotion in [12], [14], the transformation from Model node and the robot model make up a complete
(vx , vx , ω)T to W is described in (2) and (3) and param- robot. As its name suggests, the World Model node is
eters are shown in Table 2. built mainly to set up a virtual world, storing the key
  information in the simulated environment, including sta-
vx
W = A ×  vy ; (2) tuses of models, the coach information and the strategies
ω of its team.
  N ubotControl : The Nubot Control node is the core
−cos(α1 +β) −sin(α1 +β)
−d of the robot. Based on its “world”, the Nubot Control
 −cos(α
sinβ sinβ
−sin(α2 +β)  node makes decisions and sends velocity commands. With
1  2 +β)
−d 
A= × sinβ sinβ  (3)
r  
−cos(α3 +β)
sinβ
−sin(α3 +β)
sinβ −d 
 2 http://www.ros.org/core-components/
−cos(α4 +β) −sin(α4 +β)
sinβ sinβ −d 3 http://gazebosim.org/tutorials?tut=ros_installing&cat=connect_ros
standard interfaces with Gazebo, topic “VelCmd” can
control the motion of the corresponding robot model. For
every robot, the World Model node and the Nubot Control
node are unique and distinct, which comprises the basis
of the distributed strategy subsystem.
StrategyP ub : The Strategy Pub node is built for com-
munication between different robots. Because the strategy
subsystem is distributed, every robot does not know its
teammates’ strategies, which are necessary to cooperate
with its teammates. In the real world scenario, robots
share their strategies using RTDB [15]. But it is neither
feasible nor necessary for the simulation because all robots Fig. 6: The GUI of coach machine
share the same IP (Internet Protocol) address. Therefore,
the Strategy Pub node is built for the convenient and By clicking these command buttons, corresponding game
reliable communication via ROS messages. It subscribes commands are sent to robot players. The main function
to “StrategyInfo” and collects strategy information from of the display area is to visualize the positions of robot
its robots at first. Then, it fuses all strategies and publishes players, soccer ball and obstacles. In Fig 6, the yellow
new “StrategyInfo”. circles stand for robot players, and the blue circles stand
Coach4sim : The Coach4sim node is an important for obstacles, including opponent robot players. The status
component of the scene subsystem. By sending game area is used to exactly show robot players’ statuses,
commands, included in the topic “receive_from_coach”, including the position and orientation, the velocity, the
it controls the process of the game. action, the information of ball dribbling, etc.

V. Scene Subsystem B. The Automatic Referee

As aforementioned, the scene subsystem is built to As stated earlier, the coach machine is built for each
describe a specific scene. However, it is hard to understand team and it only connects with its robot players, which
the scene for the robot. Besides, to realize the confronta- makes it difficult to realize a confrontation. Therefore, the
tion, all robots agree on the process of the game. Therefore, automatic referee is proposed to not only to send game
we defines game commands, a series of commands, to commands but also to realize a confrontation conveniently.
divide the game into a series of stages, which is reasonable It is designed to judge automatically robot players’ penal-
for that a real game is usually the repetition of simple ties and send game commands to all robot players.
stages. General speaking, the definition of different stages
is consistent with rules of a specific scene. For example,
a soccer match includes stages of game-start, game-stop,
kick-off, corner-ball, penalty etc. The coach machine and
the automatic referee are able to send game commands to
robots and they are equipped with different functions.

A. The Coach Machine

The coach machine is built for each team. It aims not
only to send game commands but also to realize the Fig. 7: A graph of the automatic referee.
interaction between human and robot players. It is also A graph of the automatic referee is shown in Fig. 7. The
a state visualization tool. It displays the various states automatic referee joins in the graph as the auto_referee
of the robot, which can be used to diagnose whether the node. By subscribing to topic “/gazebo/model_states”
robot is in error, and assist in the improvement of the from /gazebo, it obtains exact statuses of all models. Then,
multi-robot collaborative strategies. The Coach Machine it is able to judge if there is a foul or a goal according
is built with three basic functions: to the rules and sends game commands to all robots by
(1) Obtaining all information of its robot players. publishing topic “receive_from_coach”.
(2) Sending game commands to robot players and
controlling the process of soccer game. VI. Interfaces
(3) Visualize information to analyze and design better As the core of Simatch, the strategy subsystem is built
multi-robot cooperation strategies. with two key interfaces, the interface with the simulation
To ensure a simple and efficient interaction between subsystem and the interface with the scene subsystem.
human and robot players, a graphical user interface (GUI) Since these interfaces are based on the ROS topic/service
is built (see Fig. 6). The GUI is divided into three areas: publishing/subscription mechanism, it enables the appli-
the command area, the display area and the status area. cation of the simulation system in different research fields.
A. Gazebo Interface The test lasts about 25.56 s and the update step is
The Gazebo interface is mainly built on the basis of about 0.015 s and results are shown in Fig. 9.
gazebo_ros_pkgs, which provides wrappers around the
stand-alone Gazebo. In Fig. 8, topic “set_model_state”
describes the desired position and velocity of models
and topic “set_link_state” describes the desired posi-
tion and velocity of links. Topic “link_states” and topic
“model_states” consist of the exact status of simulation
models. Besides, the package provides other application
programming interfaces to apply the force and the torque
on simulation models and links. These application pro-
gramming interfaces make it practical to realize various (a) (b)
motions of simulation models, even some motions are very Fig. 9: The trajectory (a) and the orientation (b) of the
complicated. Moreover, since application programming robot.
interfaces are standardized, Simatch is able to employ Two indicators are defined to measure the simulated
different models for different scenes. It supports not only robot. The first one is the position error, computed by
most of featured models from Gazebo, but also self-built the difference between the samples’ distance to the given
models from researchers such as our simulated soccer circle center and the given radius. Another one is the
robot. For some specific scenes, it is allowed to build a orientation error, representing the error between actual
brand new simulation models. On the ground of various orientation and the orientation computed by ω. After
simulation models, it is easy to modify the simulated world statistical analysis, the average position error of 1704
of Simatch for any specific scene. samples is 2.2481 cm and the average orientation error
is 0.4401 rad. The result proves that the omnidirectional
locomotion model is realized.
B. Simulation of an MSL Match
Firstly, the communication between each side of robots
is tested. In this step, though all robot model are spawned,
just a team of robots run its codes and only one coach
Fig. 8: A graph of the Gazebo.
machine is employed. With the GUI of coach machine, we
B. Scene Interface send game commands to cyan robots and observe statuses
As shown in Fig. 1, the game command is the only infor- of them. The result is shown in Fig. 10(a) and Fig. 10(b).
mation flow from the scene subsystem to the strategy sub-
system. Therefore, the scene interface only includes topic
“receive_from_coach”, which consists of M atchM ode and
M atchT ype. M atchM ode represents the current game
command and M atchT ype records the last valid game
command, and both of them are used to describe an exact
stage of a game. Though the scene interface is very simple,
it is enough to combine effectively the strategy subsystem (a) (b)
and the scene subsystem. Moreover, it makes the scene
subsystem highly independent, which makes it feasible to
design various scenes with different game commands.
VII. Validations
A. Test of the Omnidirectional Locomotion
To validate the omnidirectional locomotion, a series of (c) (d)
velocity commands are sent to a single robot and the Fig. 10: Simulation of an MSL match. (a) Simulation with
behaviors of the robot are observed. In this test, the given an active cyan team; (b) Coach for the cyan team; (c)
velocity commands make up the analytic function of a cir- Cyan’s kick off; (d) Robot players scrambling the ball.
cle. The initial status of the robot is (x, y, θ)T = (0, 0, 0)T Secondly, we run the automatic referee to control the
in the world coordinate frame. The unit of x and y is cm game process. And then, the simulated match runs by
and the unit of θ is radian. The context of topic “VelCmd” itself and the automatic referee makes correct punishment
is depicted in (4): to robots, which proves that the simulation of an MSL
(vx , vy , ω)T = (200, 0, 2)T (4) match is realized.
C. Simulation of an Encirclement MSL simulation match4 in the Chinese Robotics Compe-
To verify that it is easy to apply Simatch to simulate tition. Besides, a simulation project5 based on Simatch
other aggressive scenes, we use it to simulate an encir- has been proposed to promote the development of MSL
clement between sixteen robots, nine chasers (cyan robots) in the 2017 MSL international workshop.
rounding up another seven targets (magenta robots). The B. Future work
result is shown in Fig. 11.
Our future work will focus on two aspects. At first, we
will create more simulation models to simulate different
kinematic models, such as drones, to cater for different
application scenarios. Then, we will set up different sim-
ulation worlds and design different application scenarios
to expand the simulation system to other research fields.
References
(a) (b) [1] R. Soetens, R. V. D. Molengraft, and B. Cunha, RoboCup MSL
- History, Accomplishments, Current Status and Challenges
Fig. 11: Simulation of an Encirclement. (a) Cyan robots Ahead. Springer International Publishing, 2015.
round up magenta robots; (b) All magenta robots are [2] B. Browning and E. Tryzelaar, “Übersim:a multi-robot simu-
driven into the prison area. lator for robot soccer,” in Proc. of the 2nd Int. joint Conf. on
Autonomous agents and multiagent systems, 2003, pp. 948–949.
The simulation of an encirclement verifies that not only [3] T. Laue and S. Kai, SimRobot – a general physical robot
the number of robots can be adjusted easily but also the simulator and its application in robocup. Springer-Verlag, 2006.
[4] S. Balakirsky and Z. Kootbally, “Usarsim/ros: A combined
scene can be modified to different scenarios. Simatch can framework for robotic control and simulation,” in ASME 2012
be applied to simulate various aggressive scenes between International Symposium on Flexible Automation, 2012, pp.
multi-robot systems. 101–108.
[5] D. Beck, A. Ferrein, and G. Lakemeyer, A Simulation Envi-
VIII. Conclusion and Future Work ronment for Middle-Size Robots with Multi-level Abstraction.
Springer-Verlag, 2008.
A. Conclusion [6] W. Yao, W. Dai, J. Xiao, H. Lu, and Z. Zheng, “A simulation
system based on ros and gazebo for robocup middle size league,”
To test multi-robot collaboration strategies in MSL in IEEE International Conference on Robotics and Biomimetics,
matches, we set up Simatch based on our previous work. 2016, pp. 54–59.
The paper describes its general architecture and three [7] J. Xiao, D. Xiong, W. Yao, Q. Yu, H. Lu, and Z. Zheng, Building
Software System and Simulation Environment for RoboCup
subsystems. The simulation subsystem models the sim- MSL Soccer Robots Based on ROS and Gazebo. Springer
ulated environment and realizes basic motions of simu- International Publishing, 2017.
lation models. To improve the flexibility of robots and [8] W. Yao, Z. Zeng, X. Wang, H. Lu, and Z. Zheng, “Distributed
encirclement control with arbitrary spacing for multiple anony-
enhance the antagonistic between robots, we simulate the mous mobile robots,” in Chinese Control Conference, 2017.
omnidirectional locomotion based on the real robot. The [9] W. Yao, H. Lu, Z. Zeng, J. Xiao, and Z. Zheng, “Distributed
strategy subsystem originates from the distributed robot static and dynamic circumnavigation control with arbitrary
spacings for a heterogeneous multi-robot system,” Journal of
codes. Actually, distributed multi-robot systems have been Intelligent & Robotic Systems, pp. 1–23, 2018.
widely used in many domains. The scene subsystem is the [10] J. Ma, W. Yao, W. Dai, H. Lu, J. Xiao, and Z. Zheng,
key to simulate the confrontation between multi-robot “Cooperative encirclement control for a group of targets by
decentralized robots with collision avoidance,” in The Chinese
systems. By sending game commands to all robots, it Control Conference, 2018.
controls the process of the match. To combine these inde- [11] W. Dai, H. Lu, J. Xiao, and Z. Zheng, “Task allocation without
pendent subsystems, two convenient interfaces are built communication based on incomplete information game theory
for multi-robot systems,” Journal of Intelligent & Robotic
based on the ROS topic/service publishing/subscription Systems, no. 1, pp. 1–16, 2018.
mechanism, which make Simatch flexible and extensible [12] C. Wang, X. Liu, X. Yang, F. Hu, A. Jiang, and C. Yang, “Tra-
enough to be employed in different research fields. Then, jectory tracking of an omni-directional wheeled mobile robot
using a model predictive control strategy,” Applied Sciences,
a single robot omnidirectional locomotion test is carried vol. 8, no. 2, p. 231, 2018.
out and proves that the omnidirectional locomotion is [13] D. Xiong, J. Xiao, H. Lu, Z. Zeng, Q. Yu, K. Huang, X. Yi, and
realized. Later, a simulated MSL match between 10 robot Z. Zheng, “The design of an intelligent soccer-playing robot,”
Industrial Robot, vol. 43, no. 1, pp. 91–102, 2016.
players is simulated to verify its ability to simulate [14] Z. Li, C. Yang, C.-Y. Su, J. Deng, and W. Zhang, “Vision-based
the confrontation between multi-robot systems. Finally, model predictive control for steering of a nonholonomic mobile
we simulate the encirclement with sixteen robots, which robot,” IEEE Transactions on Control Systems Technology,
vol. 24, no. 2, pp. 553–564, 2016.
verifies the flexibility and extensibility of Simatch. [15] F. Santos, L. Almeida, P. Pedreiras, and L. S. Lopes, “A real-
To sum up, Simatch is able to simulate the highly dy- time distributed software infrastructure for cooperating mobile
namic confrontation between multi-robot systems, which autonomous robots,” in International Conference on Advanced
Robotics, 2009, pp. 1–6.
makes tests of adversarial multi-robot strategies conve-
nient and effective, and promotes studies on adversarial 4 https://github.com/nubot-nudt/simatch
multi-robot problems. Simatch has been applied to the 5 https://github.com/RoboCup-MSL/MSL-Simulator/wiki

View publication stats