Team Edinferno Description Paper for RoboCup 2014 SPL

This paper summarizes progress made by our robotic soccer team, Team Edinferno, towards our participation in the 2014 RoboCup Standard Platform League competition held in João Pessoa, Brazil. Our team made its RoboCup debut in the 2011 world cup held in Istanbul, Turkey, where we entered both in the Standard Platform League and the 2D Simulation League. In our debut at the SPL, where we entered a mostly homegrown team, we were eliminated after the first round. We returned to the 2012 Mexico City RoboCup competition, where our team reached the quarter-finals, losing to the defending champions and eventual finalists, B-Human. This version of our code leveraged the publicly available B-Human framework to provide us a software base and modules for walking and low-level vision. By 2013 we had implemented and improved on all other modules including robot behaviour, optimised kicks and goalkeeper behaviour, team level coordination/communication and probabilistic localisation. For the 2014 competition, we have worked on our software framework, implementing improvements and advances to online-generation of trajectories, localisation by better leveraging team-level information, and improved dribbling/kicking behaviours. As a team, our primary research interests are centred on issues of robot learning, especially for effective autonomous decision making and strategic behaviour through intention prediction in complex scenarios. 1 Team Introduction and Focus Team Edinferno is a team consisting of undergraduate and graduate students combined with experienced researchers from the School of Informatics at the University of Edinburgh. We come from a strong research group studying robot learning, situated within a diverse community of AI and computer science researchers the largest and best in the UK. The team leader is Dr. S. Ramamoorthy, who has extensive background in robotics and machine learning, in academia and industry. Research within our group is organized around the theme of developing autonomous decision making mechanisms in continually changing and strategically rich environments, while also leveraging our established strengths in robot control and motion synthesis. Our efforts focus on advancing specific subsystems where we bring unique research expertise and competitive advantage, using a publicly available software infrastructure framework as base. Areas where we bring novel solutions this year include improved decision making and pathplanning between kicks and dribbles; introduction of a coaching module i.e. vision-based state of play recognition and strategiser; and improvement of our probabilistic localization module with side-of-pitch disambiguation and reduction of misleading team communication. Towards continually improving the robustness of core modules, ongoing research projects explore new avenues that we aim to integrate into the fully deployed code in the near future. Continuing past research on external feature and landmark extraction from natural background images [16], we are working on finding a compromise between the accuracy and time cost for deploying an external landmark localization module to complement our current probabilistic particle filtering. Continuing past work on full-body off-line motion research [10], we are developing an on-line parametrised kick trajectory generation module to expand the range of kicking motions. Continuing research in motion strategy recognition [5], we are investigating ways to achieve opponent modelling via iterative improvements of our decision framework. We would like to acknowledge the contribution of the following Heriot-Watt ViBot program Masters students for their hard work: Jessica Abele, Mariia Dmitrieva, Isabel Schlangen, Viktor Stefanovski, Jose Luis Part, Mariela de Lucas, as well as Informatics MSc students, Kshitij Tiwari and Iris Kyranou. Fig. 1. 2014 Edinferno SPL team. Left to right: Efstathios Vafeias, Jose Luis Part, Gwendolijn Schropp, Mariela de Lucas Alvarez, Stanislav Manilov, Iris Kyranou, Svetlin Penkov, Kshitij Tiwari, Nantas Nardelli, Alejandro Bordallo and Subramanian Ramamoorthy.