memory leak?

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

memory leak?

Goutham Mallapragada
Hi Rosen,

I am trying to run a simulation(using openravepy) where I am trying to pick up a box with a dual arm robot over and over again with randomized position and size of the box.
Two things are happening:
1) The simulation crashed with MemoryError after some time. Is it possible there is a memory leak some where as I can see the memory growing gradually? I don't think it is my python script
as python has garbage collection as far as I know.
2) For certain positions of the box the simulation seems to be stuck. I attached the process to a debugger and whenever I pause the execution it always pauses
in GetCollisionInfo() line 502 of odecollision.h. The debugger cannot trace the call stack from there for some reason( I did compile openrave in RelWithDebInfo mode). The simulation  does eventually
move on but it seems to be stuck in certain places for an awful lot of time.

Any suggestions on how to track down these problems. Can I use bullet collision checker in windows?

Goutham

------------------------------------------------------------------------------

_______________________________________________
Openrave-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/openrave-users
Reply | Threaded
Open this post in threaded view
|

Re: memory leak?

Rosen Diankov
Administrator
hi Goutham,

1) the memory leak might be due to shared_pts having circular
dependencies and not releasing each other. These errors should
definitely be fixed, so if you can send a simple script to run that
reproduces the problem, we can check it out.

2) This is most likely due to the randomized planner. Randomized
planners can sometimes take a very long time, usually researchers
terminate them early and restart all over again. The reason you are
always breaking into collision loops is because 95%+ of the time spent
in a planner is in the collision checker

As for bullet, although it offers more stable tri-tri mesh checking, I
don't think the advantages are as big as thought previously, so that's
why there is no windows support yet.

rosen,

2010/4/24 Goutham Mallapragada <[hidden email]>:

> Hi Rosen,
>
> I am trying to run a simulation(using openravepy) where I am trying to pick
> up a box with a dual arm robot over and over again with randomized position
> and size of the box.
> Two things are happening:
> 1) The simulation crashed with MemoryError after some time. Is it possible
> there is a memory leak some where as I can see the memory growing gradually?
> I don't think it is my python script
> as python has garbage collection as far as I know.
> 2) For certain positions of the box the simulation seems to be stuck. I
> attached the process to a debugger and whenever I pause the execution it
> always pauses
> in GetCollisionInfo() line 502 of odecollision.h. The debugger cannot trace
> the call stack from there for some reason( I did compile openrave in
> RelWithDebInfo mode). The simulation  does eventually
> move on but it seems to be stuck in certain places for an awful lot of time.
>
> Any suggestions on how to track down these problems. Can I use bullet
> collision checker in windows?
>
> Goutham
>
> ------------------------------------------------------------------------------
>
> _______________________________________________
> Openrave-users mailing list
> [hidden email]
> https://lists.sourceforge.net/lists/listinfo/openrave-users
>
>

------------------------------------------------------------------------------
_______________________________________________
Openrave-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/openrave-users
Reply | Threaded
Open this post in threaded view
|

Re: memory leak?

Rosen Diankov
Administrator
hi goutham,

in c++, the plannerparameters structures has a _nMaxIterations fields
that controls the number of iterations the planner performs before
giving up. In the rmanipulation interfaces, you can usually specify
"maxiters" to set this parameter. In the python interfaces, this is
reflected as "maxiters"

rosen,

2010/4/27 Goutham Mallapragada <[hidden email]>:

> Rosen,
>
> Thanks for the reply. I will try to create a small script with which you can
> reproduce the problem. Meanwhile, is there any option in openrave to tell
> the planner to quit planning after certain amount of time.
>
> Goutham
>
> On Sun, Apr 25, 2010 at 12:27 AM, Rosen Diankov <[hidden email]>
> wrote:
>>
>> hi Goutham,
>>
>> 1) the memory leak might be due to shared_pts having circular
>> dependencies and not releasing each other. These errors should
>> definitely be fixed, so if you can send a simple script to run that
>> reproduces the problem, we can check it out.
>>
>> 2) This is most likely due to the randomized planner. Randomized
>> planners can sometimes take a very long time, usually researchers
>> terminate them early and restart all over again. The reason you are
>> always breaking into collision loops is because 95%+ of the time spent
>> in a planner is in the collision checker
>>
>> As for bullet, although it offers more stable tri-tri mesh checking, I
>> don't think the advantages are as big as thought previously, so that's
>> why there is no windows support yet.
>>
>> rosen,
>>
>> 2010/4/24 Goutham Mallapragada <[hidden email]>:
>> > Hi Rosen,
>> >
>> > I am trying to run a simulation(using openravepy) where I am trying to
>> > pick
>> > up a box with a dual arm robot over and over again with randomized
>> > position
>> > and size of the box.
>> > Two things are happening:
>> > 1) The simulation crashed with MemoryError after some time. Is it
>> > possible
>> > there is a memory leak some where as I can see the memory growing
>> > gradually?
>> > I don't think it is my python script
>> > as python has garbage collection as far as I know.
>> > 2) For certain positions of the box the simulation seems to be stuck. I
>> > attached the process to a debugger and whenever I pause the execution it
>> > always pauses
>> > in GetCollisionInfo() line 502 of odecollision.h. The debugger cannot
>> > trace
>> > the call stack from there for some reason( I did compile openrave in
>> > RelWithDebInfo mode). The simulation  does eventually
>> > move on but it seems to be stuck in certain places for an awful lot of
>> > time.
>> >
>> > Any suggestions on how to track down these problems. Can I use bullet
>> > collision checker in windows?
>> >
>> > Goutham
>> >
>> >
>> > ------------------------------------------------------------------------------
>> >
>> > _______________________________________________
>> > Openrave-users mailing list
>> > [hidden email]
>> > https://lists.sourceforge.net/lists/listinfo/openrave-users
>> >
>> >
>
>
>
> --
> Goutham Mallapragada, Ph.D.
> Machine Learning Scientist
> Universal Robotics, Inc.
> 2518 Smith Springs Rd
> Nashville, TN - 37217
>
>

------------------------------------------------------------------------------
_______________________________________________
Openrave-users mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/openrave-users