What is sim-to-real in robotics?

MuJoCo provides physics simulation used in robotics and related work. A simulator represents selected properties of a system. Ask which physical conditions the experiment models and which it leaves out. The fidelity needed for a result depends on the task; the name of the simulator cannot settle the question.

Look for the physical robot, transferred checkpoint or controller, calibration procedure and actual task trials. Record any fine-tuning or manual adjustments made after simulation training. Compare the claimed benefit with a relevant baseline on the same physical setup. Keep simulated trial counts separate from physical trial counts. A paper can report both without making them interchangeable.

Unspecified friction, sensor noise, latency, object variation and hardware differences can leave a transfer claim hard to interpret. These are questions to inspect, not proof that a particular method fails. Report what the source measured and leave the other fields unknown. A real-world test on one task is narrower than reliable transfer to many tasks or robot platforms.