Research and writing from the projectsim team on simulation and robot evaluation.
We weighed and tested real objects to see how closely physical properties estimated from images (mass, friction, hardness and bounce) match reality. Mass and friction estimates beat simple defaults, though a single friction coefficient per object can't capture contacts that vary by surface.
Simulated evaluation is increasingly used alongside real-world evaluation of robot policies because it is cheaper and easier to repeat; however, its value depends on how closely its outcomes track the real robot's. We test whether our reconstruction pipeline, combining metrically scaled object geometry, authored physical parameters and scene reconstruction, reduces disagreement between simulated and real robot scores relative to a default open-source recipe.
A decade after YCB gave grasping research a shared set of objects and protocols, we look at how robotics benchmarks have evolved, where they still fall short, and what simulation-ready assets make possible next.
Start turning images into sim-ready 3D at scale.
Start building