Visual and collision geometry misaligned in Gazebo 1.0

edit

I recently updated to fuerte and attempted to simulate the same robot model that I was using in Electric. It is defined in URDF and is loaded using spawn_model. Even though this model works fine in the fuerte version of RViz, it is completely broken in the new version of Gazebo (i.e. all of the collision and visual geometries are in the wrong places).

I haven't been able to diagnose the issue, but I have found a minimal example that breaks. Load the following model into Gazebo and view collision geometry. Even though I didn't modify any of the origins in the URDF, there is a huge offset between the two models. This becomes even buggier if you add non-zero <origin/>s inside the collision and geometry nodes.

I am able to simulate the PR2 fine, e.g. pr2_wg_world.world in the pr2_gazebo package, so I am not sure what I am doing wrong.

Skeleton URDF Example

<?xml version="1.0"?>
<robot xmlns:xacro="http://ros.org/wiki/xacro" name="test">
    <xacro:property name="robot_weight" value="90.0"/>

    <link name="base_link">
        <inertial>
            <mass value="${robot_weight}"/>
            <inertia ixx="1.000" ixy="0.000"  ixz="0.000"
                                 iyy="28.500" iyz="0.000"
                                              izz="34.800"/>
        </inertial>
        <collision>
            <geometry>
                <box size="0.83 0.60 0.25"/>
            </geometry>
        </collision>
        <visual>
            <geometry>
                <box size="0.83 0.60 0.25"/>
            </geometry>
        </visual>
    </link>
</robot>

Screenshot of the Problem in Gazebo (debs)

Screenshot of the Problem in Gazebo (svn)

Edits:

• Edit 1: Nothing changes if I add <origin xyz="0 0 0" rpy="0 0 0"/> to the <geometry/> elements.
• Edit 2: This may only be a problem in how Gazebo 1.0 renders the collision model. It's behaving as if the collision model is correct, even though it's rendering in the wrong position.
• Edit 3: It's broken in a different way using the SVN version of Gazebo. See above. The output of urdf2model appears to be correct, so I think it's a bug in Gazebo.
• Edit 4: I fixed the issues with the collision model by modifying how I called spawn_model. It is only buggy if I pass a non-zero offset (e.g. -z 0.5).