Frenet KF Opponent Tracker — multi_tracking.py

The more uncertain the measurement, the more R tuning dominates. σ_true is the real sensor noise (used to generate measurements); R_assumed is what the filter believes the noise is (used in the update) — the two are independent. Set R too small (over-confident) and the estimate chases measurement noise; too large (under-confident) and it is smooth but lags the truth. Watch the RMSE below for the cost. Turn LoS off and measurements stop — the prediction then coasts toward the racing-line target speed via the control input.

R-tuning presets (relative to current σtrue):
track / racing line opponent TRUE measurement (σtrue, only in LoS) KF estimate (+covariance) ego