RoboFEI-HT Team Description Paper for the Humanoid KidSizeLeague

Danilo H. Perico1, Thiago P. D. Homem1, Isaac J. da Silva1, Claudio Vilao1, Vinicius Nicassio1,Flavio Tonidandel2 and Reinaldo A. C. Bianchi1

Abstract— This Team Description Paper presents the descrip-tion of the RoboFEI-HT Humanoid League team as it stands forthe Latin American Robotics Competition 2015 in Uberlandia,Minas Gerais, Brazil. The paper contains descriptions of themechanical, electrical and software modules, designed to enablethe robots to achieve playing soccer capabilities in the environ-ment of the RoboCup Soccer Humanoid KidSize League.

I. INTRODUCTION

This paper describes the hardware and software aspectsof the RoboFEI-HT, designed to compete in the HumanoidKidSize League.

The development of the Humanoid team started in 2012,with students designing and building a humanoid robot fromscratch. Last year, the RoboFEI-HT team competed ourfirst RoboCup World Competition, held in Joao Pessoa, PB,Brazil.

At RoboCup 2014, the team stayed among the 16 topteams and in the same year, the team competed the LatinAmerican Robotics Competition (LARC 2014) and becamechampion in the LARC RoboCup Humanoid KidSize league.Fig. 1 show a game in RoboCup 2014 (1a) a game in LARC2014 (1b - an unofficial match played versus UNBeatables- NAO) and another game in LARC 2014 (1c). The officialmatches during the LARC 2014 were versus EDROM endITAndroids.

We are working on several improvements in the hardwareand software aspects of the robots and we have a team thatis able to compete and win the Latin American RoboticsCompetition.

II. ROBOT DESIGN

Our team consists of four B1 Robots. B1 Robots aredeveloped by us, but they have the mechanical parts basedon DARwIn-OP [1]. The robots will be described in thissection.

A. Hardware

B1 Robot’s can be seen in Fig. 2. B1 Robot has 20 degreesof freedom (DOF), as follows: six per leg, three per arm andtwo in the neck. The robot’s specification is depicted in TableI.

In order to improve the performance of the robots, severalstudies focused on the material stress were performed. Some

1 Department of Electrical Engineering, Centro Universitario da FEI, SaoBernardo do Campo, Brazil. rbianchi@fei.edu.br

2 Department of Computer Science, Centro Universitario da FEI, SaoBernardo do Campo, Brazil. flaviot@fei.edu.br

(a) RoboCup 2014

(b) LARC 2014

(c) LARC 2014

Fig. 1: Games played in RoboCup 2014 and LARC 2014.

research has been done to replace some metal parts by ABSparts (designed to be made in a 3D printer) in order tomaintain the strength, but with lower weight.

As a project goal, we decided to minimize the use ofelectronic parts in the robot. Our aim was to reduce all

(a) Front view

(b) Side view

Fig. 2: B1 Robot.

TABLE I: B1 Robot Characteristics

Robot B1Height 490 mmWeight 3.1 Kg

Walking speed 10 cm/sDOF 20 in total: 6 per leg,

3 per arm,2 in the neck

Type of motors Dynamixel RX-28Motors com protocol RS-485

IMU Sensor CH Robotics UM6Camera Logitech Full-HD Pro

Webcam C920Computing unit Intel NUC Core i5,

8GB SDRAM120GB SDD

Battery LiPo18.5V - 2000mAh

possible processing units and other accessory hardware,concentrating all the processing in one computer. We decidedto control the motors using the computer’s USB port. Thus,we eliminated the use of a microcontroller board, that is oftenused as an intermediate step between the computer and themotors.

As the serial communication port is not easily availableon today’s computers (especially the smaller ones), wedeveloped a USB/RS485 adapter with power supply forthe motors, so the computer can communicate with theservomotors.

With this, all the processes used by the robot, includingmotion, vision, decision, localization, planning and commu-nication are executed on the computer. Before eliminatingthe microcontroller board, that is traditionally used by mostof the teams, we conducted several tests to validate this newhardware and software architecture.

We use the UM6 ultra-miniature orientation sensor [2],featuring gyros, accelerometers, magnetic sensors to estimatethe absolute sensor orientation 500 times per second. Tocommunicate with sensors, the robot uses a USB-to-serial-adapter able to read 3 gyro axes, 3 accelerometer axes, and3 magnetometer axes.

So, this year, we improved the algorithm and our robotscan walk fast, without using a microcontroller.

B. Software

The team’s software was completely developed by ourgroup, using no software from other teams. We used ahybrid architecture, named Cross Architecture shown inFig. 3, where each one of the solid line box in the CrossArchitecture is a completely independent process for the

Fig. 3: The Cross Architecture.

computer [3], composed by a vision system, localizationsystem, decision system, planning system, communicationsystem, sense system and control system. As only oneprocessor is used, the processes were divided into 4 clustersdue to their computational cost (dashed line polygons infigure 3) and allocated in one core of the processor, asfollows:

• Cluster 1: Vision (core 0);

• Cluster 2: Localization (core 1);

• Cluster 3: Decision, Communication and Planning(core 2);

• Cluster 4: Sense and Act (IMU and Control) (core 3).

Cross Architecture is hybrid because there are someaspects of reactive paradigm and others, hierarchical. Anexample that uses a reactive paradigm is the relation betweenControl and IMU process (a type of Sense-Act) and in theother hand, process like Vision, Localization and Decisionuse hierarchical paradigm.

To communicate between the processes, Cross Architec-ture uses a blackboard, so independent processes can accessa global database. In the proposed architecture, the global

database was created using shared memory, which con-tributed to increase the velocity of the data exchange amongprocesses. To guarantee the functionality of the architecture,all shared variables are mapped before the begin of thesoftware development, so the processes can publish theirshared variables, but any other process can read this value.

III. RESEARCH INTERESTS

Our main current research interests are:

• Gait generation and optimization: usingReinforcement Learning [4], [5], Particle SwarmOptimization and Simulated Annealing;

• Stabilization Methods: using Reinforcement Learningto prevent the robot from falling down;

• Vision: using Hough Transform, Histogram of OrientedGradients (HOG) and Support Vector Machines tocreate a robust vision system, including the newRoboCup tasks of finding white ball and goalposts;

• Robot Localization: using Qualitative SpatialReasoning [6], [7], [8], [9] for self localization of therobots and for reasoning about spatial strategies;

• Multi-Robot Task Allocation: we aim to adapt asystem developed for our RoboCup Small Size Leagueteam [10];

• Spacial reasoning in multi-robot systems: using anew formalism, which we call Collaborative SpatialReasoning [11], based on Qualitative Spatial Reasoning,that can be applied on the scene interpretation frommultiple cameras and on the task of scene understandingfrom the viewpoints of multiple robots;

• Case Based Reasoning for soccer games: usingCBR together with Reinforcement Learning, to have acollection of cases that can work as set-pieces duringthe game [12], [13];

• Mechanical design of humanoid robots: to build arobot using lighter parts and new kinematic configura-tions.

As it can be seen, our research interests range from thevery bottom level of the robot construction, to the highlevel intelligent control of the team behavior. This researchhas been proved very rewarding, as we had several papersaccepted at Brazilian conferences, 2 papers accepted in theInternational Latin American Robotics Symposium [3], [14],2 papers accepted in the International RoboCup Symposium[15], [16], and 2 papers published in major journals [17],[18].

IV. CONCLUSION

In this paper we have presented the specifications ofthe hardware and software aspects of RoboFEI-HT, a teammade to compete at the Humanoid Soccer Category in theLatin American Robotics Competition (LARC 2015), heldin Uberlandia, Minas Gerais, Brazil.

ACKNOWLEDGMENT

The authors acknowledge the Centro Universitario da FEIand the Robotics and Artificial Intelligence Laboratory forsupporting this project. The authors would also like to thankthe scholarships provided by CAPES and CNPq.

REFERENCES

[1] Romela: DARwIn-OP: Open platform humanoid robot for researchand education, http://www.romela.org/main/DARwIn\_OP

[2] Pululu: Um6 ultra-miniature orientation sensor datasheet, http://www.pololu.com/file/0J442/UM6\_datasheet.pdf

[3] Perico, D.H., Silva, I.J., Vilao Jr., C.O., Homem, T.P.D., Destro, R.C.,Tonidandel, F., Bianchi, R.A.C.: Hardware and Software Aspects ofthe Design and Assembly of a New Humanoid Robot for RoboCupSoccer. In: SBR-LARS Robotics Symposium, Joint Conf. on Robotics,IEEE, p. 73-78 (2014)

[4] SUTTON, R. S. Learning to predict by the methods of temporaldifferences. Machine Learning, Hingham, MA, USA, v. 3, 1988.

[5] WATKINS, C. J. C. H. Learning from Delayed Rewards. 1989. 234 f.Tese (Doutorado) — King’s College, Cambridge, UK.

[6] COHN, A. G.; RENZ, J. Qualitative spatial reasoning. In: HARME-LEN, F. van; LIFSCHITZ, V.; PORTER, B. (Ed.). Handbook ofKnowledge Representation. [S.l.]: Elsevier, 2007.

[7] CHEN, J. et al. A survey of qualitative spatial representations. TheKnowledge Engineering Review, v. 30, p. 106–136, 10 2015. ISSN1469-8005.

[8] MORATZ, R.; NEBEL, B.; FREKSA, C. Qualitative spatial reasoningabout relative position: The tradeoff between strong formal propertiesand successful reasoning about route graphs. In: Spatial CognitionIII, Routes and Navigation, Human Memory and Learning, SpatialRepresentation and Spatial Learning. [S.l.: s.n.], 2003. p. 385–400.

[9] Fenelon Pereira, V., Gagliardi Cozman, F., Santos, P., Fernandes Mar-tins, M.: A qualitative-probabilistic approach to autonomous mobilerobot self localisation and self vision calibration. In: Intell. Systems(BRACIS), 2013 Brazilian Conf. on. pp. 157–162 (Oct 2013)

[10] Gurzoni, Jose Angelo, J., Tonidandel, F., Bianchi, R.A.C.: Market-based dynamic task allocation using heuristically accelerated rein-forcement learning. In: L. N. in Computer Science, vol. 7026, pp.365–376. Springer (2011)

[11] Santos, P., Santos, D.E.: Towards an image understanding system formultiple viewpoints. In: Brazilian Symposium on Intelligent Automa-tion (2013)

[12] Bianchi, R.A.C., Lopez de Mantaras, R.: Case-based multiagent re-inforcement learning: Cases as heuristics for selection of actions. In:19th European Conf. on Artif. Intell. IOS Press (2010)

[13] Ros, R., Arcos, J.L., Lopez de Mantaras, R., Veloso, M.: A case-basedapproach for coordinated action selection in robot soccer. Artif. Intell.173(9-10), 1014–1039 ( 2009)

[14] Vilao Jr., C.O., Silva, I.J., Perico, D.H., Homem, T.P.D., Tonidandel,F., Bianchi, R.A.C.: A Single Camera Vision System For a HumanoidRobot. In: SBR-LARS Robotics Symposium, Joint Conf. on robotics,IEEE, p. 181-186 (2014)

[15] Bianchi, R.A.C., Costa, A.H.R.: Implementing computer vision algo-rithms in hardware: an FPGA/VHDL–based vision system for mobilerobot. In: RoboCup–01: Robot Soccer World Cup V. L.N. in Artif.Intell., vol. 2377, pp. 281–286. Springer Verlag (2002)

[16] Celiberto, L.A., Ribeiro, C.H.C., Costa, A.H.R., Bianchi, R.A.C.:Heuristic reinforcement learning applied to robocup simulation agents.In: RoboCup 2007: L. N. in Computer Science, vol. 5001, pp. 220–227. Springer (2007)

[17] Bianchi, R., Martins, M., Ribeiro, C., Costa, A.: Heuristically-accelerated multiagent reinforcement learning. Cybernetics, IEEETrans. on 44(2), 252–265 (Feb 2014)

[18] Gurzoni, Jose Angelo, J., Martins, M.F., Tonidandel, F., Bianchi,R.A.C.: On the construction of a robocup small size league team.Journal of the Brazilian Computer Society 17(1), 69–82 (2011),http://dx.doi.org/10.1007/s13173-011-0028-4