[Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: Mac Mason - 2009-09-09 14:14 ```Hello list! I'm using Gazebo to test some probabilistic algorithms, and it would be very handy to have a probabilistic motion model for the simulated Pioneer 2DX. It's clear that the simulation is moving probabilistically, but (given my fairly limited understanding of Gazebo) I haven't been able to figure out where that's done, or if it lends itself to a clean parametric form at all. Another possibility would be to learn the model by just driving around, but that would require being able to query Gazebo to tell me the robot's ground-truth pose. Can somebody point me at a way to do that? Thanks! --Mac -- Julian "Mac" Mason mac@... http://www.cs.duke.edu/~mac ```
 Re: [Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: Mac Mason - 2009-09-10 20:16 ```On Sep 9, 2009, at 10:13 AM, Mac Mason wrote: > Hello list! > > I'm using Gazebo to test some probabilistic algorithms, and it would > be very handy to have a probabilistic motion model for the simulated > Pioneer 2DX. It's clear that the simulation is moving > probabilistically, > but (given my fairly limited understanding of Gazebo) I haven't been > able to figure out where that's done, or if it lends itself to a clean > parametric form at all. > > Another possibility would be to learn the model by just driving > around, > but that would require being able to query Gazebo to tell me the > robot's > ground-truth pose. Can somebody point me at a way to do that? I've poked around with Gazebo, and it looks like SimulationProxy's GetPose methods should do the right thing. However, even though they run, they seem to be lying to me (sensor readings taken at time N are not consistent with those taken at N+1, even though I'm using "ground truth" robot poses). Can anybody shed some light on this? Thanks! --Mac -- Julian "Mac" Mason mac@... http://www.cs.duke.edu/~mac ```
 [Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: Damjan Miklic - 2009-09-11 09:01 ```> Date: Thu, 10 Sep 2009 16:14:09 -0400 > From: Mac Mason > Subject: Re: [Playerstage-users] Gazebo motion model for a simulated >        Pioneer 2DX? > To: playerstage-users@... > Message-ID: <527BB312-2326-4B75-AD2B-CBB704D68FF6@...> > Content-Type: text/plain; charset=us-ascii; format=flowed; delsp=yes > > On Sep 9, 2009, at 10:13 AM, Mac Mason wrote: >> >> ... >> >> Another possibility would be to learn the model by just driving >> around, >> but that would require being able to query Gazebo to tell me the >> robot's >> ground-truth pose. Can somebody point me at a way to do that? > > I've poked around with Gazebo, and it looks like SimulationProxy's > GetPose methods should do the right thing. However, even though they > run, they seem to be lying to me (sensor readings taken at time N are > not consistent with those taken at N+1, even though I'm using "ground > truth" robot poses). > Could you describe a bit more the inconsistencies that you are getting ? I was also having some trouble with data consistency and the GetPose3d method generally seemed to be more reliable than GetPose2d. However, I was using the SVN trunk of Gazebo and was getting different results with different revisions. One revision where both GetPose3d and GetPose2d seemed to work well was r7127. damjan ```
 Re: [Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: Mac Mason - 2009-09-11 18:19 ```On Sep 11, 2009, at 5:01 AM, Damjan Miklic wrote: > Could you describe a bit more the inconsistencies that you are > getting ? I'm using the simulated laser (sicklms200) to get readings. When I take a laser reading, plot the endpoints, move the robot, take another reading, and plot those points, the points don't line up (I get wall doubling). I expect this to happen using the odometry (Position2dProxy), but not using the ground-truth robot pose. I haven't noticed any differences between GetPose3d and GetPose2d. Are we misunderstanding what the interface actually does, or is Gazebo lying to us? --Mac -- Julian "Mac" Mason mac@... http://www.cs.duke.edu/~mac ```
 Re: [Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: Mac Mason - 2009-09-15 14:40 ```On Sep 13, 2009, at 7:28 PM, Mac Mason wrote: > I'm actually using operator[], which I understand to be a wrapper > around GetRange. I'm constructing the angles manually. With further hacking around, it looks like this was a timing bug. A more-careful arrangement of my Read() calls, plus a few calls to sleep(), seem to have (mostly!) fixed it. What seems to have happened is that I was asking Gazebo for range data faster than it was actually providing it (even though the GUI had update), so I was sometimes getting old data. --Mac -- Julian "Mac" Mason mac@... http://www.cs.duke.edu/~mac ```
 Re: [Playerstage-users] Gazebo motion model for a simulated Pioneer 2DX? From: dimajk - 2009-09-18 11:54 ```Glad that you have fixed it :) And thanks for the feedback. damjan Mac Mason-2 wrote: > > On Sep 13, 2009, at 7:28 PM, Mac Mason wrote: >> I'm actually using operator[], which I understand to be a wrapper >> around GetRange. I'm constructing the angles manually. > > With further hacking around, it looks like this was a timing bug. A > more-careful arrangement of my Read() calls, plus a few calls to > sleep(), seem to have (mostly!) fixed it. > > What seems to have happened is that I was asking Gazebo for range data > faster than it was actually providing it (even though the GUI had > update), so I was sometimes getting old data. > > --Mac > > -- > Julian "Mac" Mason mac@... http://www.cs.duke.edu/~mac > > -- View this message in context: http://www.nabble.com/Gazebo-motion-model-for-a-simulated-Pioneer-2DX--tp25397453p25507469.html Sent from the playerstage-users mailing list archive at Nabble.com. ```