[ros-users] AMCL general Disuccsion

[Brian Gerkey]

hi Suraj,

Adding sonar support to amcl would be a really nice contribution.
There is some old, not-well-tested support for sonar in the underlying
library that was migrated from Player, but it has not been exposed in
the ROS node.

As for comparing performance: the goal of localization is to estimate
the robot's pose.  As such, a natural error metric is the difference
between the estimated pose and the true pose.  In simulation, this
metric is easy to compute, because the simulation can provide ground
truth.  When working with a physical system, you need some kind of
external metrology system (e.g., motion capture) as a source of ground
truth.

	brian.

On Fri, Jun 25, 2010 at 9:46 PM, Suraj Swami <suraj.g.swami at gmail.com> wrote:
> Hi Brian,
> I am Master's Student at University of North Carolina doing Individual
> project. This will help me to learn ROS in depth as well as practical
> implementation of AMCL.
> To my knowledge the AMCL driver is very complete. I could get it working in
> Player. Mainly I am trying to make filters for  sonar or might be laser and
> incorporate in the same package. I am in the impression that this package is
> written at most optimized. So after I make some changes in the method to
> weight the poses I would like to compare to the performance of this package
> to see how well the filters are doing. This is my main goal for the
> project.
>
> Hence I am trying to get the AMCL working before I make changes to it.
> Please feel free to comment and suggest on this approach.
> Actually I am still not sure how to compare the performance. Can you suggest
> of parameters that should be of concern. Or any ROS tools that will help me
> get clear picture of the performance.
>
> Thank you for all the support.
> -Suraj.
>>
>> Message: 10
>> Date: Wed, 23 Jun 2010 19:38:01 -0700
>> From: Brian Gerkey <gerkey at willowgarage.com>
>> Subject: Re: [ros-users] AMCL general Disuccsion
>> To: ros-users at code.ros.org
>> Message-ID:
>>        <AANLkTilhlZdhUXiknkqUuw2jNw0QZbRItqBeB4pvS3C5 at mail.gmail.com>
>> Content-Type: text/plain; charset="iso-8859-1"
>>
>> Hi Suraj,
>>
>> The amcl package is an implementation of the algorithm described in the
>> ProbRob book. Is there something lacking in that implementation?
>>
>>    brian.
>>
>> On Jun 21, 2010 3:12 PM, "Suraj Swami" <suraj.g.swami at gmail.com> wrote:
>>
>> Hi,
>>
>> I have some general question about AMCL. I am not sure if this the right
>> place to ask. Just that I am working on my own so I need some guidance.
>>
>> I have written Monte Carlo localization algorithm in Matlab. And I am now
>> writing it in ROS such that I make the required modification as per my
>> environment and use it for a practical robot. If it works out very good I
>> would even like to make it open source so that it is available on ros
>> repositories.
>>
>> Would you recommend me to make modification in the present amcl node or
>> write from scratch ?
>>
>> I am confused whether to use OpenCV or Geometry package for matrix and
>> maths
>> operations.
>>
>> What would be the parameters to check the efficiency of my program. I will
>> be mainly focusing on Pose Calculation time and ability to handle
>> kidnapping
>> problem. I guess these are solved problems. So there any thing this I must
>> be really concerned about.
>>
>> I am referring to book the "Probabilistic Robotics" and papers from IEEE.
>>
>> Do you recommend any specific package that would help me.
>>
>> Thank you.
>>
>> Regards,
>> Suraj Swami
>>
>>
>>
>>
>> _______________________________________________
>> ros-users mailing list
>> ros-users at code.ros.org
>> https://code.ros.org/mailman/listinfo/ros-users
>> -------------- next part --------------
>> An HTML attachment was scrubbed...
>> URL: /discuss/ros-users/attachments/20100623/ebd7971a/attachment-0001.htm
>
> _______________________________________________
> ros-users mailing list
> ros-users at code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>
>