Team-NUST Team Description for RoboCup-SPL 2014 in João Pessoa, Brazil Dr. Yasar Ayaz 1, Sajid Gul Khawaja 2, 1 RISE Research Center Department of Robotics and AI School of Mechanical and Manufacturing Engineering 2 Department of Computer Engineering, College of Electrical and Mechanical Engineering National University of Sciences and Technology, Islamabad, Pakistan yasar@smme.nust.edu.pk { sajid.gul.2009, teamnust}@gmail.com Abstract. The paper discusses the current status of the team, progress of the software development and the future work of Team-NUST. We will give a brief overview of Software architecture and its modules like vision, motion, behavior, kick, prediction and communication. Team-NUST is representing National University of Sciences and Technology (NUST) from Pakistan.
1. INTRODUCTION Team-NUST is an interdisciplinary team in its inception phase, consisting of student and researchers of National University of Sciences and Technology, Pakistan. The team was assembled in early 2013 with the aim of carrying out research in the exciting and rapidly progressing field of humanoid robots, artificial intelligence, machine vision, motion planning, and navigation in partially observable stochastic environments; and with the motivation to participate in the international event of RoboCup Standard Platforms League. The team s supervisor: Dr. Yasar Ayaz (PhD Tohoku University, Japan), is a well known name in the field of humanoid robotics. His papers on Humanoid Robot Footstep Planning, Navigation and Control have been cited by leading universities in more than 12 countries including USA, Japan, France, Germany, Hungary, South Korea, China, Canada, Iran, Croatia, Singapore, Pakistan etc. He has also been included in Top 100 Educators of the World 2013 by IBC of Cambridge and has featured in Marquis Who s Who in the World 2013 as a notable academician and researcher in the field of robotics. A list of selected relevant publications of Dr. Yasar Ayaz can be seen at Appendix A. The focus of this project is to teach NAOs how to act as a team in multi-agent cooperative environment. RoboCup-SPL provides the best opportunity to further our research. 2. TEAM MEMBERS Active members of Team-NUST include students from Department of Computer Engineering and Department of Mechanical Engineering. They are as following Khawaja Waleed Iqbal, (Team Leader) Student of B.E. Computer Engineering Alamgir Nasir, Student of B.E. Computer Engineering Dania Murad, Student of B.E. Computer Engineering Amal Rauf, Student of B.E. Computer Engineering Osama Aftab, Student of B.E. Computer Engineering Umair Hassan, Student of B.E. Mechanical Engineering Syed Hammad Ullah AlQadry, Student of B.E. Mechanical Engineering Team-NUST is currently working in RISE-LAB 1 under the supervision of Dr. Yasar Ayaz and Sajid Gul Khawaja acting as co-supervisor. RISE-LAB is a celebrated part of SMME-NUST with several publications in the field of cognitive robotics and machine intelligence with a special focus on design, control and motion planning for robotic systems including mobile robots, humanoid robots, multi-legged robots, intelligent bionics and robotic manipulators etc. 1 http://rise.smme.nust.edu.pk/
Team-NUST currently has two V4(H25) and one V2(H21) Aldebaran s NAOs and intends to buy 3 more. Webots is being used as simulator. We are looking forward to participate in team competition (Soccer), technical and drop-in player challenge. 3. SOFTWARE ARCHITECTURE The Software architecture of our system follows a modular approach where different modules work collaboratively to actuate the agents (NAOs). In our architecture, vision acts our main module and its outputs provide input to several other modules. Other modules include motion, kick, prediction, obstacle avoidance, path planning, behavior, communication. Vision has specially designed outputs; so that other modules get the required input hence they don t require pre-processing the data. We will be incorporating multi-threaded processing to make our software design more efficient and will prioritize the process based on the given state and positions of the agents. We are utilizing Aldebaran s NAOqi architecture as the basis. We are using C++ to code our modules. 4. RESEARCH Team-NUST main area of research is Artificial Intelligence and mobile robots with focus on humanoid robots and navigation. 4.1 Prediction Purpose of this module is to predict the future location of the ball and positions of other NAOs, to calculate their velocities and further use this information in decision making. Hence making it a high priority and time critical task. Currently Kalman Filter is being used for this purpose with the aim to get the precise location of the ball and robots to predict their next possible position. 4.2 Strategy Team-NUST is focusing on artificial potential field based localization strategy development. The concept is to assign potential to the object (positive and negative) to the team members and opponents. Ball has a separate potential and goal keeper has the strongest field. The strength of the potentials depends on the distance between robots (between team members of same and opposite team) and current position on the field (towards opponents goal or away from it) by minimizing noise from values of their location via Kalman Filter.
4.3 Communication In some cases vision is not enough to know the location of the team members, especially when they are behind a robot, hence while passing ball (especially backpass) can only be done by using communication modules by transmitting data packets having ID and name of the robot along with the current location of the robot. A proposed solution to the know the difference in the goal posts is by determining by the respective color of jersey worn by the goal keeper standing in front of the goal or by listening to the packets sent by the goal keeper of the team and determining the distance between sender and listeners Similar techniques are being constructed for the cooperation between team members to develop the winning strategies. 4.4 Vision Our primary focus is to make NAO understand its surroundings. To recognize ball, goal post, field lines and specially to differentiate among team members and opponents. We are using OpenCV [1] for segmentation and shape based object recognition. We will use dynamic probabilistic model to detect different objects. Purely color based analysis is not feasible due to difference of lighting condition and carpet in different location. Hence it will be very hard to classify object solely based on colors, so we will use a combined probabilistic model of color and shape based recognition and Neural networks for machine learning and pattern recognition. OpenCV has a large number of built in libraries and NAO can be made to learn the objects on the field using these libraries. In the following figures; we have shown the recognition of goal posts and the ball in the environment based on our vision techniques. Goal Post Detection Orange Ball Detection
This approach will not require any calibration for a new environment, but will automatically configure its parameters based on the new environment. Our work in the vision module will also provide help in navigation, prediction, walk and kick modules as well. By following our algorithm, NAO is able to detect blue and pink color, the color of jerseys worn during the match. Using this, NAO is capable of distinguishing between its own team mates and the opposing team members. Live video is obtained by taking frames after specific intervals (not all frames). For the consecutive frames, loops are traversed to identify the aforementioned colors. After detection of colors, Scale Invariant Feature Transform [2] will be applied to detect only the blue/pink color worn by the robots and occurrence of these colors in the rest of the world will be ignored. Similar pattern will be used for the detection of goal posts. 4.5 Kick Kick is a combination of dynamics, kinematics, joint movements and stability. Stability of the robot during kick plays a major role. For static kick, there is no feedback from the camera. Once ball is perceived and the robot is positioned accordingly relative to the ball, the kick motion starts. Keyframe based approach is used. For dynamic kick [3], NAO uses gyroscope feedback and decides for itself the next movement based on the stability of the next keyframe arrival. The working of kick module is divided in small phases. 1) Detection of goal. The image from the NAO s camera is converted to HSV space after which, a color detection algorithm is used and goal is detected. NAO then changes its position accordingly towards the goal. This phase is in process.
2) Detection of ball. NAO's head-pitch angle is changed so that it could look down to get the accurate position of the ball using the bottom camera. The ball is detected using color and shape detection algorithms. NAO then move left, right or back accordingly to adjust itself in front of the ball. 3) Kick: We have implemented different types of kicks using the keyframes strategy. There are four phases of kick as shown in the pictures. Back Kick Front Kick 4) Lean Phase: Using the current values of the angle joints, ZMP (Zero Moment Point) is calculated initially. The ZMP needs to be shifted to the center of the support polygon for the kick. So using inverse kinematics theories, new ZMP is calculated and after finding the right joint angles, the angles of knee pitch, hip pitch and ankle pitch of the robot are changed accordingly. We have made the left leg as the stable leg. The ZMP is shifted toward the center of the support polygon. 2) Raise phase: The kick foot of the robot is raised. 3) Execution phase: The robot performs the desired kick using keyframe motions. It is the main phase. 4) Return back phase: The kick foot comes back to its initial state after performing the kick. The ZMP is shifted between the two legs of the robot. The stability and the Inverse kinematics modules have been implemented. The path planning module [4] is still in process and we will be working on it to make the kicks more accurate. We will also be working more on the dynamic kick as well. 4.6 Behavior This module is mainly based on the guide lines provided by Intelligent Autonomous Robotics: A Robot Soccer Case Study [5]. Although it is not the latest study, but it provides basic techniques for A.I. development of strategy for beginners. Using bottom up approach the basic thing is an action or skill i.e. action that can be performed by an agent (robot). Skill set contains a number of skills. For mutual behavior skills of individual robots are used to form a play. For state of the game,
there is a combination of skills for multiple robots to behave cooperatively. Robots execute these maneuvers based on the conditions on the field. 5. CONCLUSION Team-NUST is a new team in RoboCup-SPL, and the only SPL-team from Pakistan. Team-NUST is very enthusiastic and fully committed to participate in RoboCup- 2014. We are hoping to learn a lot from this experience which will serve as a very strong foundation for our future research work. The aim is to make advancements in A.I. for adversarial environment for mobile robots. Team-NUST is eagerly looking forward to performing well in RoboCup-2014. 6. ACKNOWLEDGEMENTS We would like to thank National University of Sciences and Technology (NUST), Pakistan for sponsoring our RoboCup SPL research. Our inspiration comes from teams like B-Human from Bremen University, UT Austin Villa from University of Texas at Austin, NAO-Devils and several other teams who are excelling in this field and whose work we have frequently been reading. The BHuman research paper [6] is being used as basis for our kick module. We would like to thank all of them. 7. REFERENCES [1] OpenCV version 2.4.6 [2] David G. Lowe (2004) Distinctive Image Features from Scale Invariant Key Points [3] Lovish, Rahul (2013) Penalty Shootouts by Aldebaran Nao [4] Rui Ferreira, Luis Paulo Reis, Antonio Paulo Moreira, Nuno Lau (2012) Development of an Omnidirectional Kick for a NAO Humanoid Robot [5] Peter Stone (2007) Intelligent Autonomous Robotics: A Robot Soccer Case Study ISBN: 1598291270, 9781598291278 (ebook) [6] F. Wenk, T.Röfer (2013). Online Generated Kick Motions for the NAO Balanced Using Inverse Dynamics.
Appendix A List of Selected Relevant Publications of Dr. Yasar Ayaz 1. Teppei Tsujita, Atsushi Konno, Shunsuke Komuzunai, Yuki Nomura, Tomoya Myojin, Yasar Ayaz and Masaru Uchiyama, Humanoid Robot Motion Generation Scheme for Tasks Utilizing Impulsive Force, International Journal of Humanoid Robotics (IJHR), World Scientific Publishing Company, Vol. 9, No. 2, pp. 1250008-1 to 1250008-23, 2012. 2. Yasar Ayaz, Atsushi Konno, Khalid Munawar, Teppei Tsujita, Shunsuke Komizunai and Masaru Uchiyama, "A Human-Like Approach Towards Humanoid Robot Footstep Planning", International Journal of Advanced Robotic Systems (IJARS), Vol. 8, No. 4, pp. 98-109, 2011. 3. Teppei Tsujita, Atsushi Konno, Yuki Nomura, Shunsuke Komizunai, Yasar Ayaz and Masaru Uchiyama, "An Impact Motion Generation Support Software", Chapter 11, Cutting Edge Robotics 2010, Advanced Robotic Systems Journal and In-Tech Publishing Company, pp. 175-186, 2010. 4. Yasar Ayaz, Atsushi Konno, Khalid Munawar, Teppei Tsujita and Masaru Uchiyama, "Navigation Planning with a Human-like Approach", Chapter 11, Mobile Robots Navigation, Advanced Robotic Systems Journal and In-Tech Publishing Company, pp. 223-240, 2010. 5. Yasar Ayaz, Takuya Owa, Teppei Tsujita, Atsushi Konno, Khalid Munawar and Masaru Uchiyama, "Footstep Planning for Humanoid Robots Among Obstacles of Various Types,", Proceedings of IEEE-RAS International Conference on Humanoid Robots (Humanoids), pp. 361-366, Paris, December 2009. 6. Yasar Ayaz, Atsushi Konno, Khalid Munawar, Teppei Tsujita and Masaru Uchiyama, "Planning Footsteps in Obstacle Cluttered Environments," Proceedings of IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), pp. 156-161, Singapore, July 2009. 7. Yasar Ayaz, Footstep Planning Among Obstacles for Humanoid Robots, Doctoral Thesis (Robotics and Machine Intelligence), Graduate School of Engineering, Tohoku University, Japan. 8. Teppei Tsujita, Atsushi Konno, Shunsuke Komizunai, Yuki Nomura, Takuya Owa, Tomoya Myojin, Yasar Ayaz and Masaru Uchiyama, "Analysis of Nailing Task Motion for a Humanoid Robot," Proceedings of IEEE/RSJ
International Conference on Intelligent Robots and Systems (IROS), pp. 1570-1575, France, September 2008. 9. Teppei Tsujita, Atsushi Konno, Shunsuke Komizunai, Yuki Nomura, Takuya Owa, Tomoya Myojin, Yasar Ayaz and Masaru Uchiyama, "Humanoid Robot Motion Generation for Nailing Task," Proceedings of IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), pp. 1024-1029, China, July 2008. 10. Yasar Ayaz, Atsushi Konno, Teppei Tsujita, Masaru Uchiyama and Khalid Munawar, "Obstacle Stepping Over Strategy for Humanoid Robots," Proceedings of SICE System Integration Division Annual International Conference (SI), pp. 533-534, Hiroshima, December 2007. 11. Yasar Ayaz, Khalid Munawar, Muhammad Bilal Malik, Atsushi Konno and Masaru Uchiyama, "A Human-like Approach towards Humanoid Robot Footstep Planning," Proceedings of JSME International Conference on Robotics and Mechatronics (RoboMec), 1A1-B02, Akita, May 2007. 12. Yasar Ayaz, Khalid Munawar, Muhammad Bilal Malik, Atsushi Konno and Masaru Uchiyama, "Human-like Approach Towards Footstep Planning," Chapter 15, Humanoid Robots: Human-like Machines, Advanced Robotics Systems Journal and I-Tech Education and Publishing, 2007. 13. Yasar Ayaz, Khalid Munawar, Muhammad Bilal Malik, Atsushi Konno and Masaru Uchiyama, "Human-like Approach to Footstep Planning Among Obstacles for Humanoid Robots," International Journal of Humanoid Robotics(IJHR), vol. 4, no. 1, pp. 125-149, 2007. 14. Yasar Ayaz, Khalid Munawar, Muhammad Bilal Malik, Atsushi Konno and Masaru Uchiyama, "Human-like Approach to Footstep Planning Among Obstacles for Humanoid Robots," Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 5490-5495, Beijing, China, October 2006. 15. Yasar Ayaz, Autonomous footstep planning for humanoid robots, Master s Thesis (Electrical Engineering: Controls and Signal Processing), College of Electrical and Mechanical Engineering, National University of Sciences and Technology (NUST), Pakistan, 2005. 16. Yasar Ayaz, Bilal Afzal, Mannan Saeed and Saeed-ur-Rehman, "Design, Fabrication and Control of a Two-Legged Walking Robot", Proceedings Of IEEE International Workshop on Robot Motion & Control (RoMoCo), pp 73-78, Poland, June, 2004.