EuRoC Forum

[Andy]

Hi!

The scoring annex tells me, that I can get points for localizing holes in the workpiece.
But how do I communicate hole positions to the simulation without inserting the rivet?!

If my localization is within the scoring bounds, but the riveting tool can not be inserted into the object, because of collisions, how can I then call the "trigger_rivet" service?

Which brings me to the next question: I think, in a previous version of the simulation, there was no collision detection between the workpiece and the riveting tool. Now the object and the robot jump if i slightly touch the workpiece when I miss the hole by a fraction of a millimeter... This causes me headaches.

Thanks,
Andy

[ZdenekM]

For me, it also seems almost impossible to insert rivet (without collision) with current version of simulator. From my experiments it seems that arm controllers are able to achieve position accuracy about 5 mm while accuracy required for riveting is in tenths of millimeter.

[dti]

Hi Andy

The position/orientation scoring is done at the time of calling 'trigger_rivet()'. The simulator appears to read the current TCP to hole transformation and bases the scores on this. This is detailed in the Technical Annex, section 4.2 (page 18).

I am also very interested, if collisions between the tip and workpiece should infact be disabled in Gazebo.

[dti]

For me, it also seems almost impossible to insert rivet (without collision) with current version of simulator. From my experiments it seems that arm controllers are able to achieve position accuracy about 5 mm while accuracy required for riveting is in tenths of millimeter.

I completely agree. We are slightly better off having cartesian goal accuracy down to about 0.5 mm (this is roughly equivalent to a real UR5 robot) and still it is very tough. The controller is also not very good at compensating for gravity, causing grief at the start of a trajectory.

[Andy]

Good to hear, that others are struggeling too...

After lots of headaches with MoveIt (using it the first time), I think, I managed to get 0.2 mm tolerances and my limiting factor curretly is the visual detection.
Nevertheless, I will try to insert the rivet with my approch according to the hole ground truth after lunch and report back than (:

[AIS]

I'm also achieving an accuracy of 0.5-1mm on the goals of the TCP.

Really hard to achieve errors below 0.3mm...

[admin]

Dear Andy,

regarding your initial question:

The scoring annex tells me, that I can get points for localizing holes in the workpiece.
But how do I communicate hole positions to the simulation without inserting the rivet?!

Actually the accuracy of localizing the hole is not measured directly, but indirectly through the accuracy of the placement of the rivet.
This is slightly different from what is stated in the scoring annex but it was stated in that way in the technical annex since its release.

To keep fair conditions for all teams we are unfortunately not able to change this.

Best regards

[dti]

Dear Andy,

regarding your initial question:

The scoring annex tells me, that I can get points for localizing holes in the workpiece.
But how do I communicate hole positions to the simulation without inserting the rivet?!

Actually the accuracy of localizing the hole is not measured directly, but indirectly through the accuracy of the placement of the rivet.
This is slightly different from what is stated in the scoring annex but it was stated in that way in the technical annex since its release.

To keep fair conditions for all teams we are unfortunately not able to change this.

Best regards

Is the simulator working as intended with collisions being detected between the tip and workpiece?

[Andy]

Dear Andy,

regarding your initial question:

The scoring annex tells me, that I can get points for localizing holes in the workpiece.
But how do I communicate hole positions to the simulation without inserting the rivet?!

Actually the accuracy of localizing the hole is not measured directly, but indirectly through the accuracy of the placement of the rivet.
This is slightly different from what is stated in the scoring annex but it was stated in that way in the technical annex since its release.

To keep fair conditions for all teams we are unfortunately not able to change this.

Best regards

Is the simulator working as intended with collisions being detected between the tip and workpiece?

Plus 1 for this question. Even with the ground truth hole poses I can not generate a plan that is exact enough to NOT trigger a collision between rivet tool and workpiece

[admin]

Dear Andy and dti,

Is the simulator working as intended with collisions being detected between the tip and workpiece?

Yes, it is intended

Best regards