Simulation accuracy is high — and we're transparent about the gap
NVIDIA Isaac Sim uses a physics-accurate simulation engine that models real-world dynamics closely: friction, weight, sensor behavior, lighting, and spatial dimensions. When trained in a well-built simulation of your specific facility, robots transfer well to the real environment.
The sim-to-real gap
No simulation is a perfect replica of the physical world. There is always a "sim-to-real gap" — differences between simulated conditions and actual conditions. We handle this in several ways:
Domain randomization during training — We train the robot across a range of simulated conditions (lighting variation, minor layout differences, obstacle variation) so it's robust to real-world variation, not just optimized for one exact configuration.
Precise facility scanning — The more accurate our input data (point cloud scan quality), the smaller the gap. We don't compromise on scanning quality.
Post-deployment fine-tuning — During commissioning, we make any real-world adjustments needed. This is fast precisely because the major training is already done.
What this means for you practically
You'll see the robot operating in simulation before hardware is ordered. If there are gaps between simulation behavior and what your operation actually needs, we address them in software — not after hardware has shipped. By the time the robot arrives at your facility, we've already resolved the major discrepancies.