# Reproducing the results of the paper

This folder contains the code that produces the results of *Screening Ground Vehicle Controllers via Test
Randomization: Sim2Real and Sim2Sim Studies*, with the small intermediate files and result records it needs.

| Folder | Study | Details |
|---|---|---|
| [`case1/`](case1/README_case1.md) | Case 1: four path-following controllers on the 1/6-scale ART vehicle; Chrono digital twin versus 60 physical runs (Sim2Real) | [`case1/README_case1.md`](case1/README_case1.md) |
| [`case2/`](case2/README_case2.md) | Case 2: PPO goal-reaching policies; reduced-order model (ROM) versus Chrono HMMWV (Sim2Sim) | [`case2/README_case2.md`](case2/README_case2.md) |

## Quick start: all tables from the released records (a few minutes, no simulation)

```bash
# Case 1: Tables I, IV, V, sensitivity, pair tests, MPC failures (inputs in case1/data, about 10 s)
cd case1
python3.12 -m venv .venv && .venv/bin/pip install -r requirements.txt
PY=.venv/bin/python
$PY analysis/baseline_orders.py data/per_start data/nominal_sim_runs.csv data/baseline_orders.json
$PY analysis/case1_rank_analysis.py data/per_start data/physical_runs.csv data/baseline_orders.json results
$PY mpc_failures/d3_range.py data/per_start data/b01_mpc_first_tick.csv \
    results/mpc_failures.json results/mpc_failure_geometry.png
cd ..

# Case 2: Table VI and all Case 2 numbers (inputs in case2/results and case2/policies, under 2 min)
cd case2
pip install -r requirements.txt
python analyze.py --set A --pool explore
python report_explore.py
python analyze_confirm.py --set B --rom-dir results/rom_B_confirm16 --chrono-dir results/chrono_B_confirm16 \
    --primary confirm --secondary confirm16 --out results/confirm_analysis.json
python behavior_stats.py B confirm16
python posthoc_confirm.py
python report_confirm.py
python paper_numbers.py            # prints every Case 2 number of the paper; writes results/paper_numbers.json
```

These commands overwrite the released outputs in `case1/results/` and `case2/results/` with identical files.
Tables II and III (physical runs) are computed from the physical logs on GitHub (stage 2 of Case 1). The two
case READMEs give the commands of every stage, their inputs, outputs, and runtimes.

## Where each result comes from

| Paper item | Script | Output |
|---|---|---|
| Table I: TR results per offset range (S0-S4) | `case1/analysis/case1_rank_analysis.py` | `case1/results/case1_rank_analysis.json`: `offset_sweep` |
| Tables II-III: physical runs and nominal simulation | `case1/physical/physical_runs.py` | `case1/results/tables_II_III.{tex,json}` |
| Table IV: TR versus nominal simulation and Monte Carlo | `case1/analysis/baseline_orders.py`, `case1/analysis/case1_rank_analysis.py` | `case1_rank_analysis.json`: `method_comparison`, `physical_pairwise_tests` |
| Table V: ablation by channel | `case1/analysis/case1_rank_analysis.py` | `case1_rank_analysis.json`: `ablation_by_channel` |
| Case 1 sensitivity (tube, horizon, number of starts) and Hoeffding bound | `case1/analysis/case1_rank_analysis.py` | `case1_rank_analysis.json`: `tube_horizon_sensitivity`, `trial_number_sensitivity`, `hoeffding` |
| MPC failures from large heading errors | `case1/mpc_failures/d3_range.py` | `case1/results/mpc_failures.json` |
| Case 2 design exploration (goal set A) | `case2/analyze.py`, `case2/report_explore.py` | `case2/results/exploration_tau_table.json`, `exploration_report.md` |
| Table VI and the Case 2 held-out results | `case2/analyze_confirm.py`, `case2/posthoc_confirm.py`, `case2/behavior_stats.py` | `case2/results/confirm_analysis.json`, `posthoc_confirm.json`, `behavior_B_confirm16.json`; all numbers in `paper_numbers.json` |

The two case READMEs map every number in the text to its output key.

## Regenerating the records

The quick start uses the released intermediate files. To regenerate them:

- **Case 1 simulation logs:** the runner (`case1/runner/`) runs inside the Docker image named in
  `case1/README_case1.md`. The logs that it produced are in `simulation_logs/` of this dataset; stage 1
  (`case1/extract/`) turns them into the per-episode tables of `case1/data/per_start/`.
- **Case 1 physical runs:** folder `2024/PathFollowingSim2real/TestingData` of
  <https://github.com/uwsbel/sbel-reproducibility>.
- **Case 2:** `case2/train_all.sh` trains the 24 policies (about 50 min each on one CPU core); `run_rom.py` and
  `chrono_runner.py` re-run the ROM and Chrono evaluations (PyChrono). The released policies and records make
  retraining unnecessary for checking the results.

## Licenses

Code in this folder and the four checkpoints in `case2/policies/released/` are under the BSD 3-Clause license of
the original repository (`case2/policies/released/LICENSE`). Data and result records are under CC BY 4.0.