Grasping Simulation with Tacktile Sensing

Hello,

I was thinking about simulating a robotic arm grasping an object and moving it on a path. The trick is that I would need to simulate the grasped object with center of mass far away from the grasping manipulator.

With regards to it I have few questions:
-Is the simulator a good choice for grasping simulation(I know some papers has used it)
-Can I manipulate the center of the mass of the objects and will physics engine take inertia into account when solving the simulation?
-Are models of any robotics arms available in the simulator?
-Does simulation include friction and slippage?
-Could I possibly access forces working on an object(adding dummy joints could be OK, but I’d rather avoid it)

I would love if anyone could give me an answer or a reference for any of these questions, it would really help me with containing my reaserch during COVID lockdown.

All the best!

It can be done, but I think it’s important to carefully scope what you inputs and outputs you care about from the simulation.

If you care deeply about the forces between the robotic grasper and the object that’s being grasped, and you care about analyzing the stability or feasibility of that system, then I would not recommend a rigid body articulated dynamics simulator like DART. Those types of interactions are more suited to a domain-specific simulator like GraspIt!.

If you don’t care about the details of the contact forces between the grasped object and the grasper, then DART would be fine. However, the most reliable way to emulate a firm, stable grasp in DART is to create a WeldJoint between the grasped object and the grasper. Otherwise the rigid body dynamics have a very significant risk of causing the object to explode out of the grasp, unless you’re willing to dedicate a considerable amount of time to tuning your grasp controller.

Thank you for a great answer. I have just one more question.

I had a look at some papers which has used simulations and in paper titled “Deep Learning a Grasp Functionfor Grasping under Gripper Pose Uncertainty” they state to use DART because it is better then Grapit! in simulating movement in the grasp.

I would not be interested in the formation of a grasp, but rather in simulating when it fails when grasped item is swung around. Is it something that would be possible in Graspit! or better simulated in DART?

All the best,
Greg

I’ve never used GraspIt! myself, so I’m not in a position to advise on that. I only know that it’s a popular simulator among grasping researchers.

It’s certainly possible to simulate this kind of thing in DART, but doing it successfully will require very delicate control and careful configuration of friction parameters. It would also be better if there were a way to set ERP and CFM parameters, as mentioned in this issue. Those parameters can be used to make collision reactions more or less springy. When dealing with grasps, it’s important for collisions to not be too springy. Unfortunately DART’s API doesn’t offer the ability to set those parameters right now.