The NUbots Team Description Paper 2013

Brendan Annable, David Budden, Sophie Calland, Stephan K. Chalup, Shannon K. Fenn, Madison Flannery, Jake Fountain, Robert A.R. King, Alexandre Mendes, Mitchell Metcalfe, Steven P. Nicklin, Peter Turner, Josiah Walker

Newcastle Robotics Laboratory School of Electrical Engineering & Computer Science Faculty of Engineering and Built Environment The University of Newcastle, Callaghan 2308, Australia

http://robots.newcastle.edu.au

Abstract The NUbots team, from The University of Newcastle, Australia, has had a strong record of success in the RoboCup Standard Platform League since first entering in 2002. The team was also competitive within the RoboCup Humanoid Kid-Size League in 2012. The 2013 team comprises a new team leader and several new students. This paper summarizes the history of the NUbots team, describes the roles and research of the team members, gives an overview of the NUbots' robots and software system, and addresses relevant research projects within the the Newcastle Robotics Laboratory.

1 Introduction

The NUbots team, from the University of Newcastle, Australia, competed in the Four Legged League from 2002-2007, within the Standard Platform League from 2008-2011 and subsequently within the Kid-Size Humanoid league since 2012. The NUbots have had a strong record of successes, twice achieving a first place; in 2006 in Bremen, Germany, and, again in 2008 as part of the NUManoid team in Suzhou, China.

The central goal of the NUbots is to be a high performance competitive robot soccer team at RoboCup. The vision of the research projects associated with the NUbots team is to develop and program robots that can support humans not only for routine, challenging, or dangerous tasks, but also to improve quality of life through personal assistance, companionship, and coaching. Our mission is to contribute to a responsible development and application of robotics. Some of our projects therefore emphasize anthropocentric and biocybernetic aspects in robotics research [5]. This includes new aspects of human robot interaction and perception. The Newcastle Robotics Lab hosts several postgraduate and undergraduate research projects that are associated with the NUbots.

2 Commitment to RoboCup 2013

The Nubots commit to participation at RoboCup 2013 upon successful qualification. We also commit to provision of a person, with sufficient knowledge of the rules, available as referee during the competition.

3 History of the NUbots' participation at RoboCup

The NUbots team was founded in 2002 and participated for the first time at RoboCup in Fukuoka in the Sony Four-Legged League (3rd place). Since then the team has a strong history of competition and success in the RoboCup SPL/Four-Legged League, obtaining many top three placements and winning the title in 2006 and 2008.

During the previous two years the NUbots SPL code base has been ported to the DARwIn-OP platform and the majority of modules previously used on the NAO robot in the SPL were fielded for RoboCup 2012. Since then the majority of modules have undergone major revision in order to allow more effective use of the newer platform and in response to recent changes in the league rules.

4 Background of the NUbots Team Members

Josiah Walker is studying for a Doctorate of Philosophy in Computer Science in Reinforcement Learning and Robotics. He works on robot behaviour and machine learning for various NUbot systems. He is the NUbots team leader for 2013.

Brendan Annable is a 2nd year undergraduate student studying Software Engineering currently working on network infrastructure as well as a browserbased debugging environment.

Sophie Calland is a third year undergraduate student studying Computer Engineering and Computer Science. She is working on the NUbots configuration system to consolidate system parameters.

A/Prof. Stephan Chalup is the head of the Newcastle Robotics Laboratory. He is an Associate Professor in Computer Science and Software Engineering. He is one of the initiators of the University of Newcastle's RoboCup activities since 2001. His research area is machine learning and anthropocentric robotics.

Shannon Fenn is in the final semester of a Computer Engineering and Computer Science degree, who completed his Honours thesis on shape-based computer vision and optimisation, his works on machine vision with a focus on geometric methods.

Madison Flannery is a 3rd year Computer Science undergraduate currently working on improving the detection of field objects in the NUbots vision system.

Jake Fountain is a fourth year undergraduate student studying a combined degree in mathematics and science, with his main interests lying in machine learning and artificial intelligence.

Dr. Robert King is a Lecturer in Statistics at the University of Newcastle. His research focus is on flexibly-shaped distributions, statistical computing and Bayesian knowledge updating. He joined the NUbots in 2004 and has developed a special interest in RoboCup rules and refereeing.

Mitchell Metcalfe is a fourth year undergraduate student studying combined degree in Mathematics and Computer Science. He is working on the NUBots' configuration system to consolidate system parameters.

Peter Turner is technical staff in the School of Electrical Engineering and Computer Science. He is an expert on robot hardware and electronics and familiar with the Darwin-OP.

We also acknowledge the input of colleagues from the Newcastle Robotics Laboratory, team members of previous years and the Interdisciplinary Machine Learning Research Group (IMLRG) in Newcastle, Australia. Details are linked to the relevant webpages at www.robots.newcastle.edu.au. Other robotics lab members who have contributed over the past six months include:

David Budden is a recent Computer Engineering and Computer Science graduate, who completed his Honours thesis on colour-based computer vision and bipedal robotic locomotion.

Dr. Alexandre Mendes is deputy head of the Newcastle Robotics Lab. He is a Senior Lecturer in Computer Science and Software Engineering. He joined the group in September 2011 and his research areas are algorithms and optimisation.

Steven Nicklin is studying for a Doctor of Philosophy in the robotics lab. He has been working on localisation and modelling of the robot. Steve has been NUbots team leader several times and was member of the world champion teams in 2006 and 2008.

5 Hardware and Software Overview

The NUbots use the DARWIN-OP robot with footsensors. The team has four of these robots that are of the standard design with the exception of a reduced footsize which currently is introduced.

The NUbots team's major research focus is on using machine learning methods within the software systems of the robot to achieve increased performance and autonomy [8]. The current NUbots software source is available from [16] and is covered under the GPL. This includes associated toolkits for optimisation and learning on the robots. Our software platform is designed to work on multiple robotic platforms, and all of the individual modules have been designed with this in mind. The sensors and actuators are accessed using a standard format, regardless of the robot running the software [21].

The software is broken into a number of modules. The primary modules are: vision, localisation, behaviour, and motion. An overview of our software structure can be seen in Figure 1.

6 Acknowledgement of Use of Code

The NUbots Darwin-OP robots use a walk engine ported from the B-Human NAO robot walk engine [10]. We acknowledge the source of this code and have made changes only in the areas that allow it to inter-operate with our multiplatform framework.

7 Enhancements since RoboCup 2012

Since Robocup 2012, improvements have been made in the area of vision, with improved accuracy of ball detection [2], more accurate goalpost detection, and RANSAC based field-line detection. In connection with this, localisation performance has also been improved and will better deal with the challenges of same-coloured goals.

A new reinforcement learning framework has been introduced into the NUbots codebase. This framework currently governs the robot's head movement strategy, aiming to optimise localisation information. We hope to expand on this work in the future, extending our framework to solving other tasks.

The foot length of the Darwin-OP robots has also been reduced slightly in order to align with the 2013 maximum foot area rules. This is particularly relevant to locomotion and physical actions as learning systems must aim for a relatively smaller set of possible stable solutions.

Locomotion has been improved in the areas of both speed and stability thanks to the development of a new walk optimisation framework and associated new local optimisation algorithms. Embodiment in learning is also an important concept as it allows stability improvement learning at the RoboCup venue, using only the robot's senses as feedback.

8 Research Areas

Research is undertaken in the following areas at the Newcastle Robotics Laboratory.

Robot Vision

Vision is one of the major research areas associated with the Newcastle Robotics Lab. Several subtopics have been investigated including object recognition, horizon determination, edge detection, model fitting and colour classification using ellipse fitting, convex optimization and kernel machines. Recent work has resulted in a fully-autonomous method of colour look-up table generation using k-means clustering and support vector machines, as well as evaluation of colour spaces for unsupervised learning and occluded feature detection. Publications are available e.g. from [2, 3, 12, 23, 25, 11].

Localisation and Kalman Filters

Research on the topic of localisation focused on Bayesian approaches to robot localisation including Kalman Filter and particle filter based methods. We are interested in modifications of the Kalman Filter to handle non-ideal information from vision, incorporate increased information from multiple agents, and effectively utilise non-unique objects.

Development of the Robot Bear

In a collaborative effort with the company Tribotix and colleagues in design a bear-like robot (called Hykim) was developed [4]. The idea was to have a modular open platform using high quality Dynamixel servos.

Biped Robot Locomotion

The improvement of walking speed and stability has been investigated by the NUbots since several years and on different platforms: On the AIBO robot we achieved one of the fastest walks at that time by walk parameter evolution [24, 8]. On the Nao robot we improved existing walk engines by modifying the joint stiffnesses, or controller gains, [18, 19, 17]. The stiffnesses were selected through an iterative process to maximise the cost of transport. We investigated the application of Support Vector Machines and Neural Networks to proprioception data for sensing perturbations during pseudo quiet stance. Walk improvements have been primarily done via optimisation techniques [20, 22] with recent improvements to our framework for online optimisation of bipedal humanoid locomotion. The use of spiking neural networks has been trialled in simulation [26]. Prior to RoboCup 2012 the walk engine developed by the leading SPL team BHuman [10] was ported to the DARwIn-OP platform, and a variety of optimisation techniques were developed and successfully applied to improve walking speed and stability of the Darwin-OP.

Reinforcement Learning, Affective Computing and Robot Emotions

We investigate the feasibility of reinforcement learning or neurodynamic programming for applications such as motor control and music composition. Concepts for affective computing are developed in multidisciplinary projects in collaboration with the areas of architecture and cognitive science. The concept of emotion is important for selective memory formation and action weighting and continues to gain importance in the robotics community, including within robotic soccer. A number of projects in the Newcastle Robots lab already address this topic [9, 7, 13, 14, 30].

Gaze analysis and head movement behavioural learning

We investigated methods for human and robot pedestrian gaze analysis in [15, 28]. In another project we focus on detection analysis of salient regions/objects and wayfinding [1]. Recently we applied reinforcement learning techniques to optimising head movement behaviour, providing a robust algorithm by which a robot learns to choose landmarks to localise efficiently during a soccer game.

Manifold Learning

In several projects we investigate the application of nonlinear dimensionality reduction methods in order to achieve a better understanding and more precise and efficient processing of high-dimensional visual and acoustic data. [6, 27, 29, 28].

Other new projects

Much work has been focused on the underlying software architecture and external utilities to enable flexibility and extensibility for future research. Projects undertaken include improving the configurability of the software system via parameter consolidation, development of a web-based online visualisation and debugging utility and the application of software architectural principles in the area of machine vision. Some of this work is currently still in progress as part of the 2012/2013 robotics lab summer projects by new undergraduate students who have joined the team.

9 Related Research Concentrations

The Interdisciplinary Machine Learning Research Group (IMLRG) investigates different aspects of machine learning and data mining in theory, experiments and applications. Particular emphasis is put on interdisciplinary projects. The IMLRG's research areas include: Dimensionality reduction, vision processing, robotics control and learning, neurocomputing, evolutionary computation, optimisation, reinforcement learning, and kernel methods.

Links to publications can be found at the NUbots' webpage

References

1. Shashank Bhatia, Stephan K. Chalup, and Michael J. Ostwald. Analyzing architectural space: Identifying salient regions by computing 3d isovists. In Conference Proceedings. 46th Annual Conference of the Architectural Science Association (AN-ZAScA), Gold Coast, QLD, 2012.
2. D. Budden, S. Fenn, A. Mendes, and S. Chalup. Evaluation of colour models for computer vision using cluster validation techniques. In RoboCup 2012: Robot Soccer World Cup XVI, Lecture Notes in Computer Science. Springer, 2013.
3. D. Budden, S. Fenn, J. Walker, and A. Mendes. A novel approach to ball detection for humanoid robot soccer. In Advances in Artificial Intelligence (LNAI 7691). Springer, 2012.
4. S. K. Chalup, M. Dickinson, R. Fisher, R. H. Middleton, M. J. Quinlan, and P. Turner. Proposal of a kit-style robot as the new standard platform for the fourlegged league. In Australasian Conference on Robotics and Automation (ACRA) 2006, 2006.
5. S. K. Chalup and M. J. Ostwald. Anthropocentric biocybernetic computing for analysing the architectural design of house façades and cityscapes. Design Principles and Practices: An International Journal, 3(5):65–80, 2009.
6. Stephan K. Chalup, Riley Clement, Joshua Marshall, Chris Tucker, and Michael J. Ostwald. Representations of streetscape perceptions through manifold learning in the space of hough arrays. In 2007 IEEE Symposium on Artificial Life, 2007.
7. Stephan K. Chalup, Kenny Hong, and Michael J. Ostwald. Simulating pareidolia of faces for architectural image analysis. International Journal of Computer Information Systems and Industrial Management Applications (IJCISIM), 2:262–278, 2010.
8. Stephan K. Chalup, Craig L. Murch, and Michael J. Quinlan. Machine learning with aibo robots in the four legged league of robocup. IEEE Transactions on Systems, Man, and Cybernetics—Part C, 37(3):297–310, May 2007.
9. Stephan K. Chalup, Michael J. Ostwald, and Peter Walla. Academy of Neuroscience for Architecture (ANFA) 2012 Annual Conference, 21. September 2012, presentation (video), www.youtube.comwatch?v=PN9EdjErD9A, 2012.
10. Colin Graf and Thomas Röfer. A closed-loop 3d-lipm gait for the robocup standard platform league humanoid. In Enrico Pagello, Changjiu Zhou, Sven Behnke, Emanuele Menegatti, Thomas Röfer, and Peter Stone, editors, Proceedings of the Fifth Workshop on Humanoid Soccer Robots in conjunction with the 2010 IEEE-RAS International Conference on Humanoid Robots, Nashville, TN, USA, 2010.
11. N. Henderson, R. King, , and S.K. Chalup. An automated colour calibration system using multivariate gaussian mixtures to segment hsi colour space. In Proc. of the 2008 Australasian Conference on Robotics and Automation, 2008.
12. N. Henderson, R. King, and R. H. Middleton. An application of gaussian mixtures: Colour segmenting for the four legged league using hsi colour space. In RoboCup Symposium, Atlanta, July 2007, Lecture Notes in Computer Science, 2007.
13. Kenny Hong, Stephan Chalup, and Robert King. A component based approach for classifying the seven universal facial expressions of emotion. In IEEE Symposium on Computational Intelligence for Creativity and Affective Computing 2013. IEEE, accepted January 2013.
14. Kenny Hong, Stephan K. Chalup, and Robert King. Scene perception using pareidolia of faces and expressions of emotion. In IEEE Symposium on Computational Intelligence for Creativity and Affective Computing 2013. IEEE, accepted January 2013.
15. Arash Jalalian, Stephan K. Chalup, and Michael J. Ostwald. Agent-agent interaction as a component of agent-environment interaction in the modelling and analysis of pedestrian visual behaviour. In CAADRIA 2011. Circuit Bending, Breaking and Mending. The 16th International Conference of the Association for Computer-Aided Architectural Design Research in Asia, 2011.
16. J. Kulk, S. Nicklin, A. Wong, S. Bhatia, S. Fenn, D. Budden, J. Walker, and J. Reitveld. Nubots robocup code repository. https://github.com/nubots/robocup, January 2012.
17. J. Kulk and J. Welsh. Perturbation sensing using proprioception for humanoid robots. In Proceedings of the IEEE Conference on Robotics and Automation, 2010.
18. J.A. Kulk and J.S. Welsh. A low power walk for the nao robot. In Proc. of the 2008 Australasian Conference on Robotics and Automation (ACRA'2008), 2008.
19. J.A. Kulk and J.S. Welsh. Autonomous optimisation of joint stiffnesses over the entire gait cycle for the nao robot. In Proceedings of the 2010 International Symposium on Robotics and Intelligent Sensors., 2010.
20. Jason Kulk and James Welsh. Evaluation of walk optimisation techniques for the nao robot. In IEEE-RAS International Conference on Humanoid Robots, 2011.
21. Jason Kulk and James Welsh. A nuplatform for software on articulated mobile robots. In 1st International ISoLA Workshop on Software Aspects of Robotic Systems, 2011.
22. Jason Kulk and James Welsh. Using redundant fitness functions to improve optimisers for humanoid walking. In IEEE-RAS International Conference on Humanoid Robots, 2011.
23. S.P Nicklin, R. Fisher, and R.H. Middleton. Rolling shutter image compensation. In Robocup Symposium 2006, 2007.
24. M. J. Quinlan, S. K. Chalup, and R. H. Middleton. Techniques for improving vision and locomotion on the aibo robot. In Australian Conference on Robotics and Automation (ACRA'2003). ARAA (on-line), 2003.
25. M.J. Quinlan, S.P. Nicklin, N. Henderson, Fisher R., F. Knorn, S.K. Chalup, R.H. Middleton, and R. King. The 2006 nubots team report. Technical report, School of Electrical Engineering and Computer Science, The University of Newcastle, Australia, 2006.
26. L. Wiklendt, S. K. Chalup, and M. M. Seron. Simulated 3d biped walking with and evolution-strategy tuned spiking neural network. Neural Network World, 19:235– 246, 2009.
27. Aaron S. W. Wong and Stephan K. Chalup. Towards visualisation of soundscapes through dimensionality reduction. In 2008 International Joint Conference on Neural Networks (IJCNN 2008), pages 2833–2840. IEEE, 2008.
28. Aaron S. W. Wong, Stephan K. Chalup, Shashank Bhatia, Arash Jalalian, Jason Kulk, Steven Nicklin, and Michael J. Ostwald. Visual gaze analysis of robotic pedestrians moving in urban space. Architectural Science Review, 55(3):213–223, 2012.
29. Aaron S.W. Wong and Stephan K. Chalup. Sound-scapes for robot localisation through dimensionality reduction. In Jonghyuk Kim and Robert Mahony, editors, Proceedings of the 2008 Australasian Conference on Robotics and Automation (ACRA 2008). ARAA (on-line), 2008.
30. Aaron S.W. Wong, Kenny Hong, Steven Nicklin, Stephan K. Chalup, and Peter Walla. Robot emotions generated and modulated by visual features of the environment. In IEEE Symposium on Computational Intelligence for Creativity and Affective Computing 2013. IEEE, accepted January 2013.