Collision model

[tpatten]

The collision model does not correctly match the robot model. The corners of the base go beyond the cylinder of the collision model. @philkark this probably explains why we see collisions with objects because most of the time it is precisely the corners of the base that collide. @ipa-nhg can we use a prism instead of a cylinder?

[ipa-nhg]

@tpatten which part do you mean? this one: https://github.com/squirrel-project/squirrel_common/pull/243/files ? the bottom of the base?

[mzillich]

Yes, the bottom of the base. Its collision model is a cylinder, but actually it has these corners, beyond a cylinder. And these are not modelled.

[ipa-nhg]

what about use the real stl model of the base?

[mzillich]

yes we could do that. we are doing it for the robotino base itself, right?
if it does not cost much runtime

[ipa-nhg]

for the shell I only got this 3d model https://github.com/squirrel-project/squirrel_common/blob/indigo_dev/robotino_description/meshes/squirrel_assembly.stl

that we can do is also remove the robotino base stl, to reduce the RAM consumption

[mzillich]

that's a complex model. i would expect that ti impact on perforance
can we use a rectangle/blox instad od the cylinder to model the corners?

[ipa-nhg]

I cut the shell 3d model model https://github.com/ipa-nhg/squirrel_common/blob/ac8f13048d211169b7e98be3654e6985c5024f4a/robotino_description/meshes/base_shell_front_collision.stl

what do you think?

Here the Pr #256

I can draw also a rectangle, but I am afraid this will not work, it is the third model I do for the collision of the shell base.. If you need a perfect balance between a big enough collision but as small as possible I think the only solution is use the real 3d model...

But if you prefer a rectangle, please let me know the measures.

[mzillich]

@philkark is a collision model like the above ok for the planner? not too complex, run-time expenisve?

[mzillich]

anyway, yes lets try that

[tpatten]

After discussion with @philkark we think the complex mesh will be too complex and make collisions checks very slow. Can we just replace the half cylinder with a box instead?

[ipa-nhg]

yes, of course!

can you do that? or give me the measures of the box?

[tpatten]

The measure for the box is 48cm wide (we can just make it 50cm for safety). The depth and height can stay the same as the half cylinder in https://github.com/squirrel-project/squirrel_common/pull/243/files.

[ipa-nhg]

is this right??

if yes, 50cm is not enough

[tpatten]

A little bit longer so that it covers the front part of the robot model and then this would be perfect

[ipa-nhg]

better?

[tpatten]

Yes looks good. We will try it.

[ipa-nhg]

committed: #256

[mzillich]

@ipa-nhg @philkark More on collision models:

a) for the door opening the collision model is not tight enough, see the following images

b) I noticed the collision model for the robotino base (why is it red actually?) is much much more detailed than the visual model. I guess this unnecessarily slows down the planner. Can we simply use the simple mesh also for collision checking?

[ipa-nhg]

About the door. No problem, I can reduce the model. But remember that the bottom part of the real door is wider.

For the base, the real model is a collada file (with colour and textures, very heavy file and probably not supported by the planner). I can try to export the stl file of it, but I am not sure if this we be much better. Let's try!

It is red because I didn't define gazebo tags for those links, and by default gazebo spawn unknown object in red. I thought use different colours for the visual and collision can help to visualize and compare both models, but if this is not ok for you, let me know and I will add the tags.

[ipa-nhg]

For the door: #259
For the base: #258

Please test!

[tpatten]

Hi Nadia, I am still testing the base shell front collision model. We have the updated version now. Could we make it even slightly bigger? Because we are still colliding with objects. @philkark @mzillich what do you think?

[ipa-nhg]

@tpatten #260

I just added a scale factor to the URDF and didn't modify the mesh file, @tpatten you can modify the factor easily for your tests

[tpatten]

Great! Thanks that helps.