S2M2: comparison unfair due to different collision model?

[whoenig]

In S2M2 the collision model for the unicycle robot is a sphere. This makes comparison pretty unfair, because some of our examples are - in the S2M2 collision model - already in an invalid start configuration (and thus failing immediately).

We need to discuss: a) Switch to a circular model for the benchmark?, OR b) keep as is, but describe it as an inherent limitation in the paper

I am not sure if this circular model is fundamental, or an implementation detail that could be changed in principle.

[whoenig]

Discussed: we'll add a new robot type that's actually circular on our side and use that to compare to S2M2

[akmaralAW]

I have questions:

1. what will be the sphere radius ? If we want to be able to use the same environment, then it will be h / 2, where h = box width I think
2. I checked gen_p10_n1_0 example with the new robot model with sphere_radius = 0.25. Our propagation function motion hits the obstacle, while original one doesn't. I can show the video tomorrow.

[whoenig]

The only important thing is that it's consistent in all baselines, so changes are needed in robots.cpp as well as dynobench.

[akmaralAW]

what should be the spherical unicycle radius ?

[whoenig]

I would use 0.4m, so that we don't have to adjust the existing examples too much.

[whoenig]

This is fixed in the latest benchmark, where we use a spherical model throughout.