Write comments

Re: How to score accurate hole detection without inserting r

Fri 7. Nov 2014, 15:55

Okay, now one of the problems is clearly visible to me. Please have a look at the attached images: The TCP TF definition absolutely not fits the collision model. :idea:

Is this also by intention?!
Attachments
TCP_TF_2.png
TCP TF 2
TCP_TF_2.png (167.23 KiB) Viewed 276 times
TCP TF.png
TCP TF 1
TCP TF.png (238.85 KiB) Viewed 276 times

Fri 7. Nov 2014, 15:55

Re: How to score accurate hole detection without inserting r

Sat 8. Nov 2014, 12:44

Just for your information:
With slight modifications of the robot XACRO and with the ground truth hole poses I managed to insert some rivets.

Cheers,
Andy

Re: How to score accurate hole detection without inserting r

Sun 9. Nov 2014, 14:57

Hi Andy

Good work on finding this model discrepancy. I think it is what has been causing headaches for many people.

However, while it is easy to compensate for wrt. riveting the holes now we know there's an offset, scoring for hole accuracy (T2.1) will possibly be flawed as going to the correct TCP position (to score high in 2.1) will cause a collision.

The technical annex clearly states :

The tcp frame is located in the foremost point of the cylindrical rivet in the center of its circular face.
The z-axis coincides with the cylinder symmetry axis.


So a correct environment should have the error fixed.

Re: How to score accurate hole detection without inserting r

Mon 10. Nov 2014, 08:21

Dear Andy and dti,

The misalignment originated from the approximation error during the triangulation of the mesh. Sorry for not having noticed this earlier. The mesh has been correctly aligned in the current version of the simulator: 20112706nx57588/challenge-1-f3/updates-for-challenge-1-t8.html#p574

Please note that the tcp position has not changed.

The mesh-tcp alignment as it should look like:
tcp_tooltip.png
tcp_tooltip.png (13.53 KiB) Viewed 253 times


Best regards

Re: How to score accurate hole detection without inserting r

Mon 10. Nov 2014, 10:30

Hi all,

I've noticed that there is some discrepancy between the requirements of riveting in the technical annex and the intermediate evaluation. The technical annex says, the tcp should be 8 mm deep (down the z-axis of the hole frame) whereas the intermediate evaluation seems to be measuring deviation from the actual hole frame, i.e. not its 8mm-offset. Besides, /trigger_rivet doesn't work for me if the tcp is 8mm deep. Could anybody else please confirm that? What the final evaluation is going to measure?

Best,
Nikita

Re: How to score accurate hole detection without inserting r

Mon 10. Nov 2014, 11:15

Hello,

Like you have pointed, the evaluation measures the deviation between the hole frame and the TCP. No need to insert it 8mm deep.

Regards

Re: How to score accurate hole detection without inserting r

Mon 10. Nov 2014, 12:16

Dear AIS,

it is required that the rivet is inserted into the hole as stated in the technical annex:
After you know the accurate position of the hole, put the rivet into the hole. Then call trigger_rivet
service


Maybe there is an error evaluating if the rivet is accurately in place, but it has to be inserted after all.
(We will double check the evaluation code wrt this)

Best regards

Re: How to score accurate hole detection without inserting r

Mon 10. Nov 2014, 12:48

Hello,

In the last updates the system detects as "good" insertion when the TCP is placed in the hole, without any insertion.

But in any case, I would say that is almost impossible to insert the TCP so deep without any collision between robot and workpiece even having a perfect detection. Adjusting the tolerances of the path to the maximum the UR5 has a precision error of 0.5mm in the best cases (I could not get a better precision, with lower tolerances the controller can not close the loop and the movement never finishes), making it pretty hard to insert the TCP without any collisions inside the hole.

Regards

Re: How to score accurate hole detection without inserting r

Tue 11. Nov 2014, 09:01

admin wrote:Dear AIS,

it is required that the rivet is inserted into the hole as stated in the technical annex:
After you know the accurate position of the hole, put the rivet into the hole. Then call trigger_rivet
service


Maybe there is an error evaluating if the rivet is accurately in place, but it has to be inserted after all.
(We will double check the evaluation code wrt this)

Best regards


The evaluation code does indeed currently seem to be evaluating the score wrt. the surface and not the stated target of -8 mm. The following result is from a run where we do a successful rivet of ~7 mm into the hole.

Code:
position deviation [m]: 0.00704045286967
angular deviation [deg]: 0.437805481708
hole specific accuracy score: 0

Re: How to score accurate hole detection without inserting r

Tue 11. Nov 2014, 12:50

Dear all,

we found the bug in the evaluation code and will put the fixed verion online asap.

Best regards
Write comments




Bei iphpbb3.com bekommen Sie ein kostenloses Forum mit vielen tollen Extras
Forum kostenlos einrichten - Hot Topics - Tags
Beliebteste Themen: Cam, Uni, NES, Software, Mode

Impressum | Datenschutz