Re: Distribution vs centralization (Was Re: Announcement: GDHE

You wrote (in your message from Thursday 26)
 > 
 > [...]
 > > In distibuted multi-robots applications, if all robots send their
 > > position to GDHE, it becomes possible to detect some events
 > > (collisions,...) in GDHE and send them back to the robots because GDHE
 > > becomes the only centralisation point of the application. But you
 > > don't want to do that if you're working on a distributed
 > > architecture...
 > Why not? _Somewhere_ one has to have a component that has information
 > about all distributed component that need to be synchronized in one
 > way or another, isn't it?

What I have in mind here is based on our experience in the Martha
project. In this project we had a fleet of robots that are
accomplishing a set of missions in the same physical space. So they
need to cooperate both at the task level and at the geometrical
level. 

Once the task is decided, the geometrical coordination is completely
distributed : each robot has a trajectory and the knowledge of the
areas in which he needs to synchronise with others. If a robot as lost
the communication with the others he will act as a unkown moving
obstacle for the others. 

But in order to visualize what's going on, we didn't physically shut
down the communication channel. The failure was simulated and the
"rogue" robot still sended its position to GDHE. So GDHE had a
knowledge of the situation that was better than the knowledge of any
particular robot's supervisor. 

And many people who assisted to demonstrations of this projects had
difficulties to understand that the trajectory planning was done
completely on board of the robots and not in GDHE, just because GDHE
was displaying all trajectories sent by all robots.

-- 
					Matthieu