What's the data problem in warehouse robotics?
Warehouses are full of decisions that look simple and aren't. Which item goes in first, how to fit an odd-shaped order into the smallest box, how to build a pallet that won't tip, and which route through the aisles avoids people and blocked lanes.
Collecting those examples on a live floor gets in the way of operations, and the variety of items and layouts means a robot needs many examples to cover them. Teams also need someone to say which of two picks or packs was better.
Games and simulations let trainers make those decisions at volume without stopping a real warehouse. The limit is that game and sim demos capture decisions, order and paths through control inputs, not the contact physics of a real gripper, so they complement real-robot data rather than replace it.