Simulation vs. Reality: Validating Our Gazebo Environment
Comparing simulation predictions against real pool test data to calibrate our Gazebo environment for more accurate pre-competition testing.
Side-by-side comparison of Gazebo simulation and real pool test footage
Gazebo simulation (left) vs. real pool test (right)
Overview
Before committing to pool time, we test extensively in our Gazebo simulation environment. But how accurate is it? We ran the same maneuvers in simulation and in the pool to find out.
Graph comparing simulated vs actual thruster response curves
Thruster response: simulation vs. reality
Methodology
Ran identical mission scripts in Gazebo and in the pool
Recorded position, depth, and heading data from both
Compared thruster response, turn rates, and depth hold accuracy
Results
Metric
Simulation
Real World
Difference
Forward speed (m/s)
0.8
0.72
-10%
Yaw rate (deg/s)
45
38
-16%
Depth hold accuracy
±2cm
±5cm
Larger
Gate nav success rate
95%
60%
-35%
Calibration Changes
Added drag coefficients to match real-world forward speed
Increased sensor noise models (IMU drift was underestimated)
Added current simulation to model pool water movement
Key Takeaways
Simulation is optimistic — real-world success rates are lower
After calibration, forward speed and yaw rate match within 5%
Depth hold accuracy gap is mostly from sensor noise
Gate navigation gap is primarily from vision, not controls