Page 1 of 1

V-REP for grasp simulation

Posted: 29 Jan 2019, 13:29
by Robotron
Hello dear community,

For my recent project, I need a simulator that can check how good a grasp point is on a related object.
The simulation shall also be controlled by code:
  1. Spawn an arbitrary object at a defined location
  2. Test the graspability of pre-computed grasp points using a gripper. A good grasp means that the object does not slip, can lifted up and behaves stable for some seconds during the grasp.
  3. Reset and go to 1. with an other object
Is it possible to have such a routine implemented in V-REP utilizing the built-in script system?
Is it possible to realize physics-driven grasps using V-REP which are not too far away from reality and can a good grasp be detected?
Furthermore, I wonder if one can have a flying gripper (no gravity applies) that is controlled by code via a "setPosition" command for step 2. This is so I can perform the grasps in a timely manner, saving the inverse kinematics calculations when using a full robot.

Thanks in advance

Re: V-REP for grasp simulation

Posted: 29 Jan 2019, 14:35
by coppelia
Hello,

grasping is a difficult topic. You can try to design a gripper to closely model its real counterpart, but you will probably have to fight against instabilities, strange behaviours, etc. This of course also depends on the physics engine you use. Normally one would find a design that acceptably models its real counterpart, but by cheating/simplifying a bit. But this is not an option for you if you want to study the grasp.
You can move a gripper instantaneously to a specific position, but this is also something that is normally not feasable in reality. For that reason, if you try to do so with a dynamically enabled gripper (just having the base static), you will see strange effects if the jump is too large. Unless you dynamically reset all objects in the gripper before performing a larger jump (sim.resetDynamicObject).

Cheers

Re: V-REP for grasp simulation

Posted: 01 Feb 2019, 10:37
by Robotron
Thanks for your fast answer.