Scene2Real / Training
Train an aiming model
Interactive simulation
Watch the model aim
Loading the G1 view…
Demo completeAll fires out · Back at start
Model sprayTeacher aimSimulated fire
Examples seen—
Model aim—
Teacher aim—
Aim error—
Held-out comparison
Final policy results
Teaching examples—
Test positions—
Baseline mean error—
Learned mean error—
The baseline always aims forward. Errors compare each predicted angle with the geometric teacher; smaller is better. Learned coefficient: —.
| Target offset | Teacher angle | Learned angle | Learned error |
|---|
These results demonstrate the fitting workflow only; they do not establish real-world accuracy or suppression performance.