since I've discovered, just today, how ignorant I am regarding p/s internals (Radu got a taste of it in first person...sorry man for the buggy questions), I began to re-read p/s papers appeared during last years; in particular, it reminded me of this interesting and forward looking part, which interested me a lot -in terms of potential possibilities- like the first time i read it:

It would be interesting to distribute Stage’s compute load 
over a cluster of computers to support very large popula- 
tions (tens of thousands) in real timereal robots.
(from ThePlayer/StageProject: Tools for Multi-Robot and Distributed Sensor Systems, ICAR2003)

I do not know the exact improvements on last stage release (so I apologise in advance for a potentially useless question...), but if it's an issue still unaddressed I think it's something to start tackling (especially regarding Richard's Zoo project, which seems very interesting -still no code, though?- but that, I suppose, would require a feature like the aforementioned)


Mirko Bordignon - University of Padova, Italy


Il giorno 06/giu/06, alle ore 22:03, Brian Gerkey ha scritto:


Our recent overhaul of Player has given us the opportunity to do some  
new and interesting things.  I'd like to know: what do you want?

All ideas welcome, but I'm looking for relatively significant  
enhancements.  More like new transport mechanisms and less like a  
driver for the latest laser range-finder.

I'll assemble the responses into a prioritized wish list and post it  
at the project website.  I'll also incorporate the items from the SFU  
wiki on ideas for a Google Summer of Code project.

Here are a few ideas to get things going (feel free to comment on them):

- New transport layers, getting away from the client-server model.   
Candidates include CORBA, ACE, Reid Simmons' IPC...

- Standard implementations of multi-robot algorithms.  Whereas now we  
have implementations of single-robot algorithms like AMCL and VFH, we  
could implement multi-robot SLAM, auction-based task allocation, etc.

- Support for structured "client-to-client" communications.  Right  
now, this is totally ad hoc, with everybody reinventing the wheel.   
We could take ideas from the old lifomcom driver and new relay driver  
to provide more fully-featured communication links among control  
programs (which, depending on the transport mechanism in use, will  
not necessarily be "clients")

- Standard tasks to try in simulation, with instrumentation to  
provide automatic evaluation.  Richard's going in this direction with  
the Zoo.


Playerstage-users mailing list