File size: 5,852 Bytes
0c3dd1d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 | #!/usr/bin/env python3
"""Case 1 baseline orders for Table IV: nominal simulation and Monte Carlo evaluation on the deployment paths.
For each path P1-P3 and metric (lateral, heading) a baseline predicts the controller order by ascending run MAE:
* M-NOM: one nominal Chrono run per path and controller (batch B03, tag nominal_role = primary, controller phase 0);
* NOM-archived-2024: the nominal simulation logs of Tables II-III (physical/physical_runs.py -> nominal_sim_runs.csv);
* RF-A: Monte Carlo, start pose jittered within the spread of the physical starts, random controller phase,
20 runs per path and controller (batch B04); score = mean run MAE over the 20 runs;
* RF-B: as RF-A plus actuation delay 0.10 s, steering bias 0.013 and pose noise 0.02 m / 0.01 rad (batch B05).
Runs of the Monte Carlo batches are paired by scenario_id; a scenario that lacks a controller (or has a non-finite
MAE) is dropped for every controller. NN-HD column: ctrl_label nnhdx175 for P1/P2, nnhdx150 for P3. Order ties keep
the label order MPC, NN-MPC, PID, NN-HD (stable sort).
Usage: python baseline_orders.py PER_START_DIR NOMINAL_SIM_RUNS_CSV OUT_JSON
PER_START_DIR contains B03_nominal/, B04_rf_a/, B05_rf_b/ with fullpath_part*.csv.gz (extract stage).
"""
import csv
import glob
import gzip
import json
import os
import sys
import numpy as np
csv.field_size_limit(1 << 24)
LABELS = ['mpc', 'nnmpc', 'pid', 'nnhd']
PATHS = ['P1', 'P2', 'P3']
METRICS = ['lateral', 'heading']
METRIC_COL = {'lateral': 'mae_lat_m', 'heading': 'mae_head_rad'}
MATCHED = {'P1': ['mpc', 'nnmpc', 'pid', 'nnhdx175'], 'P2': ['mpc', 'nnmpc', 'pid', 'nnhdx175'],
'P3': ['mpc', 'nnmpc', 'pid', 'nnhdx150']}
DESC = {'M-NOM': 'nominal Chrono simulation, one run per path and controller (B03, phase 0)',
'NOM-archived-2024': 'nominal simulation logs of Tables II-III (one run per path and controller)',
'RF-A': 'Monte Carlo, start jitter and random phase, 20 runs per path and controller (B04)',
'RF-B': 'Monte Carlo, start jitter, random phase and disturbances, 20 runs per path and controller (B05)'}
def fnum(v):
return float(v) if v != '' else np.nan
def load_fullpath(d, batch_dir):
rows = []
for f in sorted(glob.glob(os.path.join(d, batch_dir, 'fullpath_part*.csv.gz'))):
with gzip.open(f, 'rt', newline='') as fh:
rows += [r for r in csv.DictReader(fh) if r['kind'] == 'fullpath']
return rows
def ordering(scores):
"""LABELS sorted by ascending score (stable; ties keep input order)."""
return [LABELS[i] for i in np.argsort(np.asarray(scores, dtype=float), kind='stable')]
def nominal(rows):
out = {}
for p, labels in MATCHED.items():
for m in METRICS:
vals = []
for lab in labels:
sel = [r for r in rows if r['path_key'] == p and r['ctrl_label'] == lab]
vals.append(fnum(sel[0][METRIC_COL[m]]) if len(sel) == 1 else np.nan)
out[(p, m)] = np.array(vals)
return out
def monte_carlo(rows):
out, n_sets = {}, {}
for p, labels in MATCHED.items():
sub = [r for r in rows if r['path_key'] == p]
for m in METRICS:
cell = {}
for r in sub:
if r['ctrl_label'] in labels:
key = (r['scenario_id'], r['ctrl_label'])
if key in cell:
raise RuntimeError(f'duplicate row for {key}')
cell[key] = fnum(r[METRIC_COL[m]])
scens = sorted({s for s, _ in cell})
complete = [s for s in scens if all((s, l) in cell for l in labels)]
complete = [s for s in complete if all(np.isfinite(cell[(s, l)]) for l in labels)]
M = np.array([[cell[(s, l)] for l in labels] for s in complete], float).reshape(len(complete), len(labels))
out[(p, m)] = M.mean(0)
n_sets[p] = len(complete)
return out, n_sets
def main():
d, nom_csv, out_json = sys.argv[1:4]
preds, info = {}, {}
b03 = [r for r in load_fullpath(d, 'B03_nominal') if r['tag_batch'] == 'B03' and r['tag_nominal_role'] == 'primary']
preds['M-NOM'] = nominal(b03)
arch = {}
for r in csv.DictReader(open(nom_csv)):
c = {'MPC': 'mpc', 'NN-MPC': 'nnmpc', 'PID': 'pid', 'NN-HD': 'nnhd'}[r['controller']]
arch.setdefault(r['path'], {}).setdefault('lateral', [None] * 4)[LABELS.index(c)] = float(r['run_mae_lateral_m'])
arch[r['path']].setdefault('heading', [None] * 4)[LABELS.index(c)] = float(r['run_mae_heading_rad'])
preds['NOM-archived-2024'] = {(p, m): np.array(arch[p][m], float) for p in PATHS for m in METRICS}
for arm, bdir, tag in (('RF-A', 'B04_rf_a', 'B04'), ('RF-B', 'B05_rf_b', 'B05')):
preds[arm], info[arm] = monte_carlo([r for r in load_fullpath(d, bdir) if r['tag_batch'] == tag])
res = {'_doc': 'Baseline predicted orders per path and metric (ascending run MAE). predicted_scores are in the '
'order MPC, NN-MPC, PID, NN-HD (P1/P2: nnhdx175, P3: nnhdx150 for the Chrono batches).'}
for name, pr in preds.items():
res[name] = {'description': DESC[name]}
if name in info:
res[name]['n_runs_per_controller'] = info[name]
for p in PATHS:
for m in METRICS:
pt = pr[(p, m)]
if not np.all(np.isfinite(pt)):
raise RuntimeError(f'{name} {p} {m}: missing runs')
res[name][f'{p}_{m}'] = {'predicted_ordering': ordering(pt), 'predicted_scores': [float(v) for v in pt]}
json.dump(res, open(out_json, 'w'), indent=1)
for name in preds:
print(name, {t: '>'.join(v['predicted_ordering']) for t, v in res[name].items() if isinstance(v, dict) and 'predicted_ordering' in v})
if __name__ == '__main__':
main()
|