Jeremy Wyatt Professor of Robotics and Artificial Intelligence Intelligent Robotics Laboratory Centre for Computational Neuroscience and Cognitive Robotics School of Computer Science University of Birmingham Edgbaston Birmingham B15 2TT +44 (0)121 414 4788 +44 (0)121 414 4281 (fax) +44 (0)121 415 8722 (Claire Benson, IRLab Administrator) Email for Claire Benson: irlabpa @ cs.bham.ac.uk) Email for teaching and admin: jlw @ cs.bham.ac.uk Email for research only

News

We have just started our new FP7 project PacMan on robot manipulation of objects.
We've just made some great hires! Ales Leonardis and Michael Mistry are joining the IRLab, taking our strength to five faculty. I'm really excited about our new arrivals.
Have a look at our video (IJCAI 11 video finalist) on probabilistic planning in robot object search here . This is really neat work joint with other labs in the CogX project in which we show how a robot can make search much faster and more reliable by incorporating commonsense knowledge about the world, and reasoning explicitly about its uncertainty about the state of the world, and how that can change.

Research Interests

I'm interested in a number of problems, all of which are motivated by the same scientific goal: studying general architectures and methods for learning and reasoning in autonomous agents, especially those with bodies. My interests are broad. I have worked on the exploration-exploitation problem in reinforcement learning, the problem of managing diversity in committees of learning machines, cognitive architectures for intelligent robotics, learning of predictions in robot manipulation, planning and learning of information gathering strategies in robots (e.g. in AUVs, in processing of images, in gaze control, or in object search), and on the use of physics knowledge in prediction and estimation in vision. There are relevant publications on these topics and many others below. My current interests are all based around the need for robots to understand their surroundings, and to be able to extend that understanding by themselves, and with others.

I am a member of the Intelligent Robotics Laboratory. The lab has about twenty five researchers. I'm also the Project Coordinator for the FP7 funded project PacMan. My funders include, or have included: the European Commission, EPSRC, NERC, the Royal Society, the Leverhulme Trust, The British Council, the Department of Trade and Industry, and AWM. Thanks to all of them for the generous support they provide.

Publications

Birmingham PhD theses in Machine Learning

Here are some great PhD theses to come out of Birmingham in Machine Learning. This includes two winners of the BCS/CPHC Distinguished Dissertation Competition. This is the highest honour that can be given to a doctoral thesis in computer science in the UK. Enjoy:

• Adrian Hartley, Variable State Methods for Learning with Hidden State , 2005. .pdf . This thesis shows the difficulties and challenges to be found in learning from reinforcement in environments with hidden state. It develops a number of Utile Distinction methods for reinforcement learning in unknown POMDPs.
• Gavin Brown, Diversity in Neural Network Ensembles , 2004. .pdf . This thesis was one of the winners of the BCS/CPHC Distinguished Dissertation Competition 2004. It shows how to precisely quantify diversity in regression ensembles, and how this can be expressed in terms of a decomposition of the error function for the ensemble.
• Stuart Reynolds, Reinforcement Learning with Exploration , 2003, .pdf , .ps.gz . This presents a number of new algorithms for reinforcement learning that are more efficient in terms of time and space, and quality of learning in continuous spaces.
• Tim Kovacs, Strength or Accuracy? Credit Assignment in Learning Classifier Systems. 2002, now available from Springer-Verlag, 2003. ISBN: 1-85233-770-2. This was one of the winners of the 2002 BCS/CPHC Distinguished Dissertation competition. In it Tim shows how different learning classifier systems work and explains their behaviour in terms of whether they are driven by accuracy or reward.

PhD Opportunities in the Robot Learning lab

Right now I'm primarily interested in recruiting PhD students with excellent mathematical and coding skills to work on aspects of the CoSy project. This is a four year, multi-site European project to study the scientific problems in building Cognitive Robots. I will consider students for other topics. Please note the following before mailing me:

• I am not accepting any more applications for internships for 2008. A competition will be announced for internships for the IRlab for the summer of 2009. The competition will open in November 2008, and details will appear on the IRlab page. Pleaes note two things. Applications prior to this will thus be ignored with no reply. Sorry! Second, I use internships as a way to identify good PhD applicants. If you want to study for a PhD in the USA and not the UK, then there is no point in entering.
• If you are applying from within the EU you should have a first degree, and preferably a Masters, and be in the top 5% of your graduating class. Typically you should have a first if you are applying from a UK institution.
• If you are looking for us to fund you, or to support a funding application on your behalf, then you must be in the top 1% of your graduating class, and preferably have some publications under your belt.
• If you are applying from the USA, or a country that uses a GPA scoring system, you should be looking at 3.5 GPA or better.
• If you are applying from Iran, or a country with a similar educational system, you should be looking at an average of 17 for a Bachelor's degree, and higher for a Master's degree.
• I am happy to be contacted by email, but you should attach a one page CV in pdf format detailing education, grades and publications. You should also write a two page summary of how your proposed research interests fit with those of our lab. If you send a non-specific enquiry I will just direct you back to this page. I get 50-60 informal PhD and internship enquiries each year, and I'm afraid I can't reply to them all quickly. Sorry if you've mailed and not heard from me yet.

In general I am happy to supervise PhD students in any area of robot learning, reinforcement learning, Bayesian or other probabilistic approaches to learning, behaviour based robotics, learning in computer vision, evolutionary robotics, multi-robot learning, reinforcement learning, or sequential decision making. I'll also consider supervising more general work in intelligent mobile robotics. If you have an idea then please e-mail me.

If you're interested in applying to do a PhD in this school you should also look at the School's web page http://www.cs.bham.ac.uk/ and follow the link labelled: How to apply to be a Research Student here. This includes pointers to an online application form and information about funding opportunities, fees, etc. A general page for all admissions information about the School is available.

Current Collaborators

• CogX Project
  • Markus Vincze, Michael Zillich, TU Wien.
  • Patric Jensfelt, Danica Kragic, KTH Stockholm.
  • Geert-Jan Kruijff, DFKI Saarbrucken.
  • Bernhard Nebel, Michael Brenner, Albert Ludwig University, Freiburg.
  • Ales Leonardis, Danijel Skocaj, University of Ljubljana, Slovenia.
  • Richard Dearden, Aaron Sloman, Nick Hawes, Rustam Stolkin, Charles Gretton, IRLab University of Birmingham.
• AFDA Project
  • Richard Dearden, Juhan Ernits, Birmingham.
  • Gwyn Griffiths, Miles Pebody, Southampton.
• Learning Visual Routines Project, UKIERI
  • Ravi Balaraman, IIT-Madras
  • Richard Dearden, Birmingham
• ADEPT Project
  • Gavin Brown, University of Manchester.
  • Tim Kovacs, University of Bristol.
• CoSy Project
  • Aaron Sloman on the CoSy project (Birmingham site, general CoSy site).
  • Nick Hawes , Cognitive Architectures for Cognitive Robots.
  • Somboon Hongeng , Learning about hum an actions from observation.
  • Michael Zillich, vision for learning about object manipulation.
  • Marek Kopicki, Learning about objects and their affordances via robotic manipulation.
• PhD Students
  • Noel Welsh, population based approaches to reinforcement learning.
  • Zeyn Saigol , autonomous science.
  • Funlade Sunmola, model transfer in reinforcement learning.
  • Damien Duff, Physics informed model based visual tracking.
  • Ismail Bhula, Learning the meaning of utterances about actions.
  • Veronica Rios, Learning about deformable objects.
  • Jose Varela.
• Other faculty members in Birmingham
  • Richard Dearden, reinforcement learning, autonomous science.
  • Peter Hancox, Learning the meaning of utterances about actions.
• Outside Birmingham
  • Gavin Brown: Learning in Committee Machines.
  • Yoshiyuki Matsumura, Evolutionary robotics.
  • Panrasee Ritthipravat from the Field Robotics Unit, Bangkok, Thailand , Knowledge Sharing in Reinforcement Learning.

Teaching

• Timetable
• Intelligent Robotics
• Introduction to AI
• Neural Networks Web Site.

Undergraduate and MSc projects on offer

I'm now full for 2008-09.

Travelling Plans

• Oct 18-23, Dagstuhl Seminar.
• Oct 27-31, CogX review, Saarbrucken.
• December 14-15, Honda Research Institute, Offenbach.
• January (Dates TBC), IIT Madras, Chennai.