Pose Estimation for Robotic Soccer Players in the Context of RoboCup - PowerPoint PPT Presentation

MIN Faculty Department of Informatics Pose Estimation for Robotic Soccer Players in the Context of RoboCup Judith Hartfill University of Hamburg Faculty of Mathematics, Informatics and Natural Sciences Department of Informatics Technical Aspects of Multimodal Systems November 28, 2017 J. Hartfill – Pose estimation 1 / 21

Outline Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References 1. Motivation 2. RoboCup RoboCup Soccer Leagues 3. Pose Estimation Pose Estimation Approaches Particle Filter Problems 4. Approaches in Robocup Soccer Humanoid Kid Size League Standart Platform League 5. Summary 6. Conclusion and Perspectives J. Hartfill – Pose estimation 2 / 21

Motivation Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References Knowing own pose is essential for decision making. How can a robot know its pose on the field? 1 1 http://clipart-library.com/images/8iGb5XKbT.jpg J. Hartfill – Pose estimation 3 / 21

RoboCup Competitions Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ International competitions ◮ Since 1996 ◮ 500 teams ◮ Several leagues 2 2 http://www.robocup2014.org/wp-content/uploads/2014/04/RCfed_high_M_Transp.png J. Hartfill – Pose estimation 4 / 21

RoboCup Industrial Leagues Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References 3 3 http://robohub.org/robocup-video-series-industrial-league/ J. Hartfill – Pose estimation 5 / 21

RoboCup Rescue Leagues Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References 4 4 http://www.robocup2009.org/21-1-robocup%20rescue.html J. Hartfill – Pose estimation 6 / 21

RoboCup@Home Leagues Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References 5 5 https://ispr.info/2011/08/01/robocuphome-2011-when-the-home-help-is-a-robot/ J. Hartfill – Pose estimation 7 / 21

RoboCup Soccer Leagues Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References 6 6 https://www.robocupgermanopen.de J. Hartfill – Pose estimation 8 / 21

RoboCup Soccer Leagues Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Humanoid Leagues ◮ Several sizes ◮ Only humanoid sensors ◮ Humanoid dimensions ◮ Adapted FIFA rules ◮ Standart Platform League ◮ NAO ◮ not humanoid 7 Win against FIFA World Cup champion in 2050 7 https://robocup.informatik.uni-hamburg.de/wp-content/uploads/2017/07/P1100771-1.jpg J. Hartfill – Pose estimation 9 / 21

Pose Estimation Approaches Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References Pattern Matching Visual Compass [3] Particle Filter [4] [6] ◮ Visual map ◮ Probabilistic ◮ Least-squares method ◮ Histogram linear regression ◮ Multiple sensor problem inputs 9 10 8 8 https://upload.wikimedia.org/wikipedia/commons/b/b0/Linear_least_squares_example2.svg 9 http://alife-robotics.co.jp/homepage2018/members2017/icarob/data/html/data/OS_pdf/OS12/OS12-4.pdf 10 http://networks.ece.mcgill.ca/sites/default/files/1.png J. Hartfill – Pose estimation 10 / 21

Particle Filter Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References J. Hartfill – Pose estimation 11 / 21

Particle filter Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References J. Hartfill – Pose estimation 12 / 21

Problems in Humanoid Kid Size League Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Odometry hard ◮ Bad vision ◮ Computationally limited ◮ Symmetry of the field ◮ Other robots occluding view ◮ ... J. Hartfill – Pose estimation 13 / 21

Approaches in Humanoid Kid Size League Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Reminder: Odometry hard ◮ Rhoban [2]: ◮ 3D Particle filter ◮ Magnetometer ◮ Field boarders and goals posts ◮ Foot pressure sensors ◮ Action model less erroneous ◮ Visual observations scored 11 11 https://www.robocuphumanoid.org/qualification/2017/22ee18648e39f3f656609d932ab6ccaa70a66929/ Rhoban_Fooball_Club_Humanoid_KidSize_regularanddrop_in_2017_TDP.pdf J. Hartfill – Pose estimation 14 / 21

Approaches in Humanoid Kid Size League Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Reminder: Bad vision ◮ ZJU[5] ◮ Particle filter with sensor resetting ◮ Input noisy ◮ Propability of particles low ◮ Replace some particles with noisy ones J. Hartfill – Pose estimation 15 / 21

Approaches in Standart Platform League Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Improvement: Communication ◮ Camellia Dragons [1] ◮ Observer view robot observing field ◮ Sharing information via WiFi ◮ Resampling particles with additional information ◮ Not natural like in usual soccer 12 Camera image True perspective image Pose estimation 12 http://alife-robotics.co.jp/homepage2018/members2017/icarob/data/html/data/OS_pdf/OS12/OS12-4.pdf J. Hartfill – Pose estimation 16 / 21

Summary Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Hardware ◮ Foot pressure sensors: Better data for particle filter ◮ Software ◮ Sensor resetting: Escape from bad estimates ◮ Observer view: Use all capacities J. Hartfill – Pose estimation 17 / 21

Conclusion and Perspectives Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References ◮ Particle filter popular and reliable ◮ Workarounds for bad sensor data ◮ Hardware improvements useful ◮ Communication becoming more important ◮ Better computers/ sensors J. Hartfill – Pose estimation 18 / 21

Conclusion and Perspectives Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References J. Hartfill – Pose estimation 19 / 21

References Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References [1] Yo Aizawa, Takuo Suzuki, and Kunikazu Kobayashi. Improvement of robot’s self-localization by using observer view positional information, 2017. The 2017 International Conference on Artificial Life and Robotics (ICAROB 2017), Jan. 19-22, Seagaia Convention Center, Miyazaki, Japan. [2] Julien Allali, Louis Deguillaume, Remi Fabre, Loic Gondry, Ludovic Hofer, Olivier Ly, Steve NGuyen, Gregoire Passault, Antoine Pirrone, and Quentin Rouxel. Rhoban football club: Robocup humanoid kid-size 2016 champion team paper . Springer Berlin Heidelberg, Berlin, Heidelberg, 2016. [3] Peter Anderson and Bernhard Hengst. Fast Monocular Visual Compass for a Computationally Limited Robot , pages 244–255. Springer Berlin Heidelberg, Berlin, Heidelberg, 2014. J. Hartfill – Pose estimation 20 / 21

References (cont.) Motivation RoboCup Pose Estimation Approaches in Robocup Soccer Summary Conclusion and Perspectives References [4] S. Lenser and M. Veloso. Sensor resetting localization for poorly modelled mobile robots. In Proceedings 2000 ICRA. Millennium Conference. IEEE International Conference on Robotics and Automation. Symposia Proceedings (Cat. No.00CH37065) , volume 2, pages 1225–1232 vol.2, 2000. [5] Mei WenXing, Pan Yusu, Peng Bo, Jiang ChaoFeng, Liu Yun, and Xiong Rong. Zjudancer team description paper, 2017. RoboCup 2017 Team Description Paper Humanoid Kid-Size League. [6] Thomas Whelan, Sonja Stüdli, John McDonald, and Richard H. Middleton. Efficient Localization for Robot Soccer Using Pattern Matching , pages 16–30. Springer Berlin Heidelberg, Berlin, Heidelberg, 2012. J. Hartfill – Pose estimation 21 / 21