It has become a common practice to use simulation to generate large databases of good grasps for grasp planning in robotics research. However, the existence of a generic simulation context that enables the generation of high quality grasps that can be used in several different contexts such as bin-picking or picking objects from a table, has to our knowledge not yet been discussed in the literature.
In this paper, we investigate how well the quality of grasps simulated in a commonly used “generic” context transfers to a specific context, both, in simulation and in the real world.
We generate a large database of grasp hypotheses for several objects and grippers, which we then evaluate in different dynamic simulation contexts e.g., free floating (no gravity, no obstacles), standing on a table and lying on a table.
We present a comparison on the intersection of the grasp outcome space across the different contexts and quantitatively show that to generate reliable grasp databases, it is important to use context specific simulation.
We furthermore evaluate how well a state of the art grasp database transfers from two simulated contexts to a real world context of picking an object from a table and discuss how to evaluate transferability into non-deterministic real world contexts.