Simulation & Data
Robots have no internet-scale corpus to learn from. This chapter covers how the field manufactures data — teleoperation, simulation, human video — and what it takes for behavior learned in one place to transfer to another.
§Articles in this chapter
1 articleThe sim-to-real gap: why simulated robots fail on hardware
The sim-to-real gap is the performance lost when a simulation-trained policy runs on a real robot. What causes it and the techniques used to close it.
§Terms from this chapter
Full glossary →Domain randomization
Training a policy in simulation while randomly varying physics and visual parameters, so that the real world looks like just one more variation it has already seen.
Sim-to-real gap
The drop in performance when a policy trained in simulation runs on a physical robot, caused by differences between simulated and real physics, sensing and timing.
Teleoperation
A human controlling a robot remotely in real time. It is used both to collect demonstration data for training and to recover robots in the field when autonomy fails.