sweep_apogee.py¶

Grid-search Kalman filter tuning parameters against recorded flights.

What it does¶

sweep_apogee.py iterates over a small grid of (q_vel, r_meas, debounce) values and, for each combination, runs every recorded flight in the target directory through process_real_flight() from sim_flight_kalman.py. For each combination it records, per flight, the signed delta between the filter-detected apogee and the ground-truth apogee computed from the log.

A configuration is rejected if:

the filter fails to detect an apogee on any flight, or
the filter fires more than 50 ms before the ground-truth apogee on any flight (no early fires).

Surviving configurations are sorted by worst-case lag across flights and the top 15 are printed.

Usage¶

The script takes no command-line arguments — the flight directory and search grid are currently hard-coded at the top of the file:

FLIGHT_DIR = AURORA_DIR / "flight_logs/2026-04-18"

grid = list(itertools.product(
    [1.5, 2.0, 3.0, 4.0, 6.0],    # q_vel
    [2.0, 4.0, 6.0, 8.0, 12.0],   # r_meas
    [1, 2],                       # debounce
))

Edit those two blocks to point at a different flight capture or to widen the search space.

Run with:

python3 tools/sweep_apogee.py

Expected output:

Sweeping 50 configs

Top 15 configs (lowest worst-case lag, no early fires):
 q_vel     r  db    f1_Δ    f2_Δ   worst
   2.0   4.0   2   +0.08   +0.12   +0.12
   ...

f1_Δ, f2_Δ — per-flight lag in seconds (positive = late, negative = early).
worst — worst-case lag across flights used as the ranking key.

Input layout¶

The referenced FLIGHT_DIR must contain:

flights.influx — telemetry dump (InfluxDB line protocol).
state_audit — state machine transition audit log.

Both can be produced by sim_fetch.py or harvested from a real flight.

Requirements¶

Python 3.10+
numpy
Must be runnable from the aurora/tools directory so the import of sim_flight_kalman resolves.