Re: RT-MBDyn

Herman Bruyninckx writes: 

> On Fri, 12 Mar 2004, Pierangelo Masarati wrote: 
> 
>> Since you've always been very friendly to our
>> work, which is strongly related to yours, let
>> me announce you that RT-MBDyn, i.e. RTAI-enabled
>> MBDyn for real-time multibody dynamics simulation,
>> has been released in beta:
> Thanks for the announcement! 
> 
> Maybe the Orocos community could benefit from a two sentences summary
> about what exactly are the "realtime" features of a simulation? In
> what respect does it differ from a "fast" simulator?

Sure. 

The simulation is scheduled in hard real time; it can be scheduled
periodically, or controlled by other events, e.g. mailboxes,
semaphores and so, which, thanks to netrpc and rtnet, can be generated
by remote processes.  This way, the simulator entirely replaces the
real system inside a simulated hardware-in-the-loop environment. 

In the cases that are being simulated right now, mostly for development
purposes, we usually address the simulation of industrial robots; the
control is performed by a separate process, generated from simulink or
scicos via RTAILab.  The processes interact by means of RTAI mailboxes. 

You can really find more details on the RT-MBDyn site, which is
maintained by Matteo Martegani. 

I'm glad you asked about the differences between RT-MBDyn and a fast
simulator.  Now we can say that MBDyn is a fast simulator; we're
dedicating a considerable amount of time in making it as fast as
possible to allow the real-time simulation of large models at
reasonable sampling rates.  The difference is in the tight
scheduling to allow interaction with reality.  The directions we're
considering for performances are:
 - dedicated sparse solvers; usually sparse solvers are optimized for
 space rather than for speed; they save time by reducing the amount
 of overall operations, but memory consumption is the real objective;
 we are working on the implementation of sparse solvers optimized
 for speed (the "naive" solver that is in the released beta);
 - parallelization; the currently released version can perform
 multithreaded element jacobian and residual assembly; however,
 since in many problems assembly is only a portion of the solution
 time, this didn't get much speed improvement for the small problems
 that are interesting for real-time; we're working on the
 parallelization of the "naive" solver, which should scale pretty well;
 - general code cleanup; we already squeezed a lot of unnecessary
 or non-optimized operations out of MBDyn; we can say it's some
 20/50% faster than before, depending on the type and size
 of the problem. 

Pierangelo. 

Dr. Pierangelo Masarati               | voice: +39 02 2399 8309
Dip. Ing. Aerospaziale                | fax:   +39 02 2399 8334
Politecnico di Milano                 | mailto:[email protected]
via La Masa 34, 20156 Milano, Italy   | http://www.aero.polimi.it/~masarati