Rapid iteration and testing what you'd never dare test physically
Long before any control algorithm touches real hardware, simulation is where it gets built and refined. A team can write a new controller, run it in simulation, watch it fail, adjust it, and rerun it within minutes, a cycle that would take hours or days if it required resetting and re-running physical hardware each time. This tight iteration loop is what makes it feasible to explore many different approaches, tune parameters extensively, and catch obvious bugs, all before committing any physical trial time to an idea, which is a limited and expensive resource by comparison.
Simulation is equally valuable for large-scale automated testing of edge cases that would be impractical or outright dangerous to test on real hardware. Want to know how a warehouse robot behaves when a person steps directly into its path at the exact moment it's rounding a corner at full speed, or how a legged robot recovers from having one leg suddenly lose torque mid-stride, or how a drone responds to a sudden gust while carrying an unbalanced payload? These are exactly the scenarios a team wants confidence in before deployment, and exactly the scenarios too risky, expensive, or simply too rare to reliably reproduce on demand with physical hardware. In simulation, thousands of variations of each scenario can be run automatically and the results checked for failures systematically.