Benchmarks

Replay throughput

Run benchmark and export artifacts:

python benchmarks/benchmark_replay.py

Artifacts:

• benchmarks/artifacts/replay_benchmark.csv
• benchmarks/artifacts/replay_summary.md

Customize durations, filter backends, and output location:

python benchmarks/benchmark_replay.py --durations 30 120 300 --filter-types ekf ukf --output-dir benchmarks/artifacts

EKF vs UKF comparison report

Generate an accuracy/consistency/speed comparison across fixed scenarios:

python benchmarks/compare_filters.py --scenarios nominal dropout outlier --duration 60 --filter-types ekf ukf --output-dir benchmarks/artifacts

Artifacts:

• benchmarks/artifacts/filter_comparison.csv
• benchmarks/artifacts/filter_comparison.md

Included metrics:

• Position RMSE (rmse_position_m)
• Velocity RMSE (rmse_velocity_mps)
• ANIS (anis_mean)
• ANEES (anees_mean)
• Throughput (throughput_eps)

The Markdown report also includes per-scenario takeaway bullets comparing best accuracy vs best speed.