search

DavidPL1 / assembly_example

Example code for interaction with our assembly simulation ICRA 2023 challenge
7 stars 1 forks source link

Some unusual problems about the screwing task #25

Open Director-of-G opened 1 year ago

Director-of-G commented 1 year ago

Hi, we encountered two unusual problems when performing the screwing task.

  1. There is inter-penetration between the gripper and the assembly components (i.e., screw and nut) when we exert force to insert. The following two figures show this case. _cgi-bin_mmwebwx-bin_webwxgetmsgimg__ MsgID=1452448688244373243 skey=@crypt_35da0724_1b4d11b9ac4bc7c4be60f38fd72c2c10 mmweb_appid=wx_webfilehelper _cgi-bin_mmwebwx-bin_webwxgetmsgimg__ MsgID=8501044387141320538 skey=@crypt_35da0724_1b4d11b9ac4bc7c4be60f38fd72c2c10 mmweb_appid=wx_webfilehelper We also provide two pieces of videos to show this problem. (video1 and video2)
  2. It is possible to have the screw or nut not on the table when the simulation started, as shown in the figure below. This problem is similar to this one. _cgi-bin_mmwebwx-bin_webwxgetmsgimg__ MsgID=822343414854087400 skey=@crypt_35da0724_1b4d11b9ac4bc7c4be60f38fd72c2c10 mmweb_appid=wx_webfilehelper Thanks for your reply!
DavidPL1 commented 1 year ago
  1. There is inter-penetration between the gripper and the assembly components (i.e., screw and nut) when we exert force to insert. The following two figures show this case.

We are aware of that problem.

This is due to the softness of the solver constraints in the physics calculation. In this case this means that the solver finds a solution where penetration causes less violation of the constraints than generating the negative acceleration needed to stop the penetrating object. I.e. the force you apply with the gripper is too high for the current solver parameters. In our tests we found that you should have no such problems if you use a grasp force up to 5 N. We are still working on finding a better solution to this.

But given the challenge deadline approaching fast, I don't want to fiddle too much with solver parameters of the physics that could have major impacts on simulation behavior. If the penetration in your case is also prevented by limiting the maximum force exerted by the gripper this would be the preferred solution right now.

  1. It is possible to have the screw or nut not on the table when the simulation started, as shown in the figure below.

Thanks for reporting this. I'll look right into it

abhishek47kashyap commented 1 year ago

screw or nut not on the table

We ran into this as well

Screenshot from 2023-04-28 22-07-19

fpatzelt commented 1 year ago

I investigated the issue of screws and nuts not being on the table. I found that in 13 of 1000 trials the screw or nut disappeared from the table in the first few seconds of simulation. In 9 cases the nut spawned in collision with the table and contact forces in the first timestep sent the nut flying. In the remaining 4 cases a screw with a wide but short top part spawned close to the edge of the table, toppled over, and fell down.

For all observed failure cases, I found similar initial configurations that are stable. So this should not affect the variety of available training scenarios in a meaningful way. We will remove these failure cases from the final evaluation in order not to negatively influence the scores of the participants.