Saturday, June 21, 2008

Refining Masters Topic

One thing I have come to learn very quickly is that what they say about masters (and research in general) being a continuous refinement of your research topic cannot be more true. If you had told me two months ago that my topic would be thinned down to what it is today I simply would not have believed you!

When I first approached my supervisor, Anet, about joining the RoboCup team I saw it as just that, an opportunity to help in putting together an intelligent system for a robotic soccer team to compete in RoboCup. And I was going to be doing the vision and localisation. My oh my, how that vision (excuse the pun) has changed.

First of all, I am no longer doing any vision-related work so my reliance on Patrick Marais as co-supervisor has fallen away. I dropped it because, to do anything novel to any degree (while masters does not require this, they try get it from you here) I had to concentrate on a narrower topic, which in my case was localisation.

Then lets look at RoboCup. Well the initial aim was to ship a prototype for the RoboCup tournament in July. If I was to have done that though, I simply would have had insufficient time to do enough reading and that would result in me wasting a lot of time essentially being a code monkey hacking away at something that would certainly not be novel and, given more time to read up on and design, would be fairly trivial. Then there's also the major problem that we simply have not yet received any robots yet!!! This just makes development of anything on the robots a lot harder and frankly I'd rather wait until we get them before doing any hacking as we might realise things we cannot know without seeing the robots. Then there's the fact that one of our team members in Austria, Christof Rath, is already working on basic localisation for July.

So we've realised that we don't get much (any?) credit from RoboCup directly. The traditional (for want of a less offensive term) folks of this world don't see RoboCup as being real research so we have to target it as a more general solution for robotics in general and tag RoboCup as simply a testbed. It's a very neat testbed though, as everything is already laid out so there's a lot less of this topic refinement I'm talking about in this post. Most masters projects typically start out weakly defined with huge scope and can even change completely over the course of the project. We simply don't have that problem, which is great as we can focus from much earlier on.

Another area of refinement, one which I briefly touched on above, is that of supervisors. Anet Potgieter was always my main supervisor from the get go as she started the local RoboCup initiative. Anet's research field is agents (subset of AI). She's so busy though: over 20 graduate students, 10 honours students and the director of a company!! Then we needed a co-supervisor in mechanical engineering: I chose Stephen Marais who's doing his PhD in SLAM (Simultaneous Localisation and Mapping: a problem more difficult than localisation on its own) and he's been great as he's always in and his office is right in our lab, even though he's really on sabbatical. Then I started out with Patrick Marais as the computer vision guru at UCT, but since I dropped that part I no longer have him as co-supervisor. Then finally, just a week ago I approached one of our newer lecturers Hanh Le and she jumped at the opportunity to co-supervise me. She's viscuous, but in a positive way. She refuses to accept students unless she knows they are top students and looks far ahead, much further than most other supervisors do. Still early to comment too much, but to give you an idea she gave me a wad of papers 15 cm tall at our second meeting! And for completeness, my second reader is Edwin Blake.

So anyway, the reason I started out with this post was that I am at the final stages of writing up my research proposal. It's done, just needs to go through final checks. The deadline is six months after starting, I've almost got it out after two so I'm already ahead of schedule. :-D The aim is to have my proposal presentation on Wednesday 2 July so if you're in the area and would like to attend (if you're not a UCT student you should ask before coming as attendance might be restricted, I'm not 100% sure), watch my blog for further details. The aim is to put my presentation together with that of Andre Scholtz who's doing the collaborative intelligence for RoboCup.

No comments:

Post a Comment