298 lines
15 KiB
Python
298 lines
15 KiB
Python
"""P4 全向闭环端到端验收：10/16/0 源、边界源、预算耗尽与故障注入。
|
|
|
|
对应 docs/B题建模分析与实施计划.md P4 行验收门槛（自建部分）：
|
|
- 10/16 源场景全部清除；16 源触发数量上界停止；
|
|
- 近圆域边界源不漏搜、可清除；
|
|
- 持续复扫 [用户决策 2026-09-11]：找全后只复扫未清除频道直到虚拟上限；
|
|
- 每步计时（实机 + 比赛虚拟耗时分项）逐步留档可汇总；
|
|
- 虚拟时刻逐步复算与服务器响应一致；
|
|
- 现实预算耗尽（找全前）退出记为未完成，不与成功混记；
|
|
- 网络故障重试（幂等回放）不污染闭环结果。
|
|
|
|
SMALL_CAP 把虚拟限时压到 60k 秒：持续复扫在数轮内触及安全余量，
|
|
避免测试扫满 360k 秒上限（等价行为，仅缩短复扫轮数）。
|
|
"""
|
|
|
|
import unittest
|
|
from dataclasses import replace
|
|
from itertools import count
|
|
|
|
from mcm26b.config import ProblemConfig
|
|
from mcm26b.coverage import coverage_points
|
|
from mcm26b.evaluation import recompute_virtual_time
|
|
from mcm26b.planner import OmnidirectionalLoop
|
|
from mcm26b.protocol.client import SimulatorClient
|
|
from mcm26b.sim import (LocalSimTransport, LocalSimulator,
|
|
ScriptedFaultTransport, Source)
|
|
from mcm26b.sim.scenes import random_omni_scene
|
|
|
|
CONFIG = replace(ProblemConfig(),
|
|
search=replace(ProblemConfig().search,
|
|
bearing_rounding_margin_deg=0.005))
|
|
SMALL_CAP = replace(CONFIG,
|
|
field=replace(CONFIG.field, max_virtual_duration_s=60_000.0))
|
|
|
|
|
|
def _run_scene(sources, fault_script=None, clock=None, config=SMALL_CAP):
|
|
simulator = LocalSimulator(sources, robot_id="T001", config=config)
|
|
transport = LocalSimTransport(simulator)
|
|
if fault_script is not None:
|
|
transport = ScriptedFaultTransport(transport, fault_script)
|
|
kwargs = {"clock": clock} if clock is not None else {}
|
|
client = SimulatorClient("T001", config=config, transport=transport, **kwargs)
|
|
loop = OmnidirectionalLoop(client, config)
|
|
summary = loop.run()
|
|
report = recompute_virtual_time(client.log.action_events())
|
|
return summary, report, simulator
|
|
|
|
|
|
class LoopAcceptanceTests(unittest.TestCase):
|
|
def test_ten_sources_all_cleared(self):
|
|
sources = random_omni_scene(seed=5, count=10)
|
|
summary, report, simulator = _run_scene(sources)
|
|
truth = sorted(s.channel for s in sources)
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "virtual_time_cap")
|
|
self.assertEqual(sorted(summary["cleared_channels"]), truth)
|
|
self.assertEqual(len(summary["absent_channels"]), 10)
|
|
self.assertEqual(summary["unresolved_channels"], [])
|
|
self.assertIsNotNone(summary["accounted_virtual_time_s"],
|
|
"找全证明时刻必须留档")
|
|
self.assertLessEqual(summary["accounted_virtual_time_s"],
|
|
summary["virtual_time_s"])
|
|
self.assertGreater(summary["measures"], 49 * 20,
|
|
"找全后应继续复扫未清除频道（扫到限时为止）")
|
|
self.assertEqual(summary["anomalies"], [], "守恒性或证书不应被打破")
|
|
self.assertEqual(report["mismatches"], [])
|
|
self.assertEqual(summary["virtual_recompute_mismatches"], 0)
|
|
self.assertEqual(sorted(simulator.cleared), truth)
|
|
self.assertEqual(summary["exit_reason"], "user_exit")
|
|
self.assertLess(summary["virtual_time_s"],
|
|
CONFIG.field.max_virtual_duration_s)
|
|
|
|
def test_sixteen_sources_stop_by_count_bound(self):
|
|
sources = random_omni_scene(seed=6, count=16)
|
|
summary, report, simulator = _run_scene(sources)
|
|
truth = sorted(s.channel for s in sources)
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "sixteen_cleared")
|
|
self.assertEqual(sorted(summary["cleared_channels"]), truth)
|
|
self.assertEqual(report["mismatches"], [])
|
|
self.assertEqual(sorted(simulator.cleared), truth)
|
|
|
|
def test_empty_field_proves_absence_everywhere(self):
|
|
summary, report, _ = _run_scene([])
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "virtual_time_cap")
|
|
self.assertEqual(summary["cleared_channels"], [])
|
|
self.assertEqual(summary["absent_channels"], list(range(1, 21)))
|
|
self.assertGreaterEqual(summary["measures"], 49 * 20,
|
|
"缺席证明要求全网格逐频道检测")
|
|
self.assertIsNotNone(summary["accounted_virtual_time_s"])
|
|
self.assertEqual(report["mismatches"], [])
|
|
|
|
def test_boundary_source_close_to_domain_edge(self):
|
|
"""半径 1790m、接收半径 1000m 的源：覆盖保证的最近网格点必须发现它。"""
|
|
source = Source(channel=13, x=1790.0, y=0.0, r_eff_m=1000.0)
|
|
summary, _, simulator = _run_scene([source])
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(sorted(summary["cleared_channels"]), [13])
|
|
self.assertEqual(sorted(simulator.cleared), [13])
|
|
|
|
def test_source_on_grid_vertex(self):
|
|
source = Source(channel=20, x=-1400.0, y=700.0, r_eff_m=1500.0)
|
|
summary, _, _ = _run_scene([source])
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(sorted(summary["cleared_channels"]), [20])
|
|
|
|
def test_real_budget_exhaustion_exits_cleanly(self):
|
|
"""现实预算在找全前耗尽：立即退出并记为未证明完成。"""
|
|
ticks = count(0.0, 200.0) # 每次查询时钟前进 200s，迅速耗尽 1200s 预算
|
|
sources = random_omni_scene(seed=8, count=4)
|
|
summary, _, _ = _run_scene(sources, clock=lambda: next(ticks))
|
|
self.assertFalse(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "real_budget_exhausted")
|
|
self.assertEqual(summary["exit_reason"], "user_exit")
|
|
|
|
def test_fault_injection_during_coverage(self):
|
|
"""覆盖中途注入读超时与断连：幂等回放保证闭环结果不变。"""
|
|
sources = random_omni_scene(seed=9, count=3)
|
|
script = [None, None, None, "timeout_after", None,
|
|
"drop_before", None, ("status", 500), None, None]
|
|
summary, report, simulator = _run_scene(sources, fault_script=script)
|
|
truth = sorted(s.channel for s in sources)
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(sorted(summary["cleared_channels"]), truth)
|
|
self.assertEqual(report["mismatches"], [])
|
|
self.assertEqual(sorted(simulator.cleared), truth)
|
|
|
|
|
|
class StepTimingTests(unittest.TestCase):
|
|
"""每步计时 [用户决策 2026-09-11]：实机耗时 + 比赛虚拟耗时分项。"""
|
|
|
|
def test_steps_recorded_with_virtual_breakdown(self):
|
|
sources = random_omni_scene(seed=5, count=2)
|
|
summary, report, _ = _run_scene(sources)
|
|
steps = summary["steps"]
|
|
self.assertEqual(steps[0]["action"], "enter")
|
|
self.assertEqual(steps[-1]["action"], "exit")
|
|
self.assertTrue(all(step["real_elapsed_ms"] >= 0 for step in steps))
|
|
first_measure = next(step for step in steps if step["action"] == "measure")
|
|
self.assertEqual(first_measure["step_virtual_s"],
|
|
first_measure["move_s"] + first_measure["switch_s"]
|
|
+ first_measure["action_s"])
|
|
totals = summary["step_totals"]
|
|
self.assertEqual(totals["count"], len(steps))
|
|
# 虚拟分项之和 = 虚拟总时长（比赛口径：移动+切频道+检测+清除）
|
|
self.assertAlmostEqual(
|
|
totals["virtual_move_s"] + totals["virtual_switch_s"]
|
|
+ totals["virtual_measure_action_s"] + totals["virtual_clear_action_s"],
|
|
totals["virtual_total_s"], delta=0.5)
|
|
self.assertAlmostEqual(totals["virtual_total_s"],
|
|
summary["virtual_time_s"], delta=0.5)
|
|
self.assertGreaterEqual(totals["real_elapsed_total_s"], 0.0)
|
|
# 清除成功一步 5s、未发现 3s：步级记录可复核
|
|
clear_step = next(step for step in steps if step["action"] == "clear")
|
|
self.assertIn(clear_step["action_s"], (3.0, 5.0))
|
|
self.assertEqual(clear_step["switch_s"], 0.0, "/clear 不切频道")
|
|
|
|
def test_avg_locate_clear_time_metric(self):
|
|
"""题面口径：平均定位清除时间 = 虚拟总时长 / 被清除个数。"""
|
|
sources = random_omni_scene(seed=5, count=2)
|
|
summary, _, _ = _run_scene(sources)
|
|
cleared = len(summary["cleared_channels"])
|
|
self.assertEqual(summary["avg_locate_clear_time_s"],
|
|
round(summary["virtual_time_s"] / cleared, 3))
|
|
|
|
|
|
class PersistentSearchTests(unittest.TestCase):
|
|
def test_persistence_disabled_stops_when_resolved(self):
|
|
"""persistent_search=False：回到"证明找全即退出"的基线行为。"""
|
|
config = replace(CONFIG, search=replace(
|
|
CONFIG.search, persistent_search=False, bearing_rounding_margin_deg=0.005))
|
|
summary, _, _ = _run_scene(random_omni_scene(seed=5, count=2),
|
|
config=config)
|
|
self.assertTrue(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "all_resolved")
|
|
self.assertLessEqual(summary["measures"], 49 * 20 + 60)
|
|
|
|
def test_persistence_resweeps_only_uncleared_channels(self):
|
|
"""复扫只测未清除频道：已清除频道不再消耗检测。"""
|
|
sources = random_omni_scene(seed=5, count=2)
|
|
summary, _, _ = _run_scene(sources)
|
|
cleared = set(summary["cleared_channels"])
|
|
resweep_channels = {step["channel"] for step in summary["steps"]
|
|
if step["index"] > 0
|
|
and step["virtual_time_s"]
|
|
> summary["accounted_virtual_time_s"]
|
|
and step["action"] == "measure"}
|
|
self.assertTrue(resweep_channels)
|
|
self.assertFalse(resweep_channels & cleared,
|
|
"已清除频道不得在找全后被复扫")
|
|
|
|
def test_virtual_budget_stops_before_simulator_cap(self):
|
|
"""计划层在虚拟安全余量处停下，不触发模拟器 accepted=false 兜底。"""
|
|
sources = random_omni_scene(seed=5, count=2)
|
|
summary, report, _ = _run_scene(sources)
|
|
self.assertEqual(summary["stop_reason"], "virtual_time_cap")
|
|
field = SMALL_CAP.field
|
|
self.assertLessEqual(summary["virtual_time_s"],
|
|
field.max_virtual_duration_s)
|
|
self.assertGreaterEqual(summary["virtual_time_s"],
|
|
field.max_virtual_duration_s
|
|
- SMALL_CAP.search.virtual_reserve_s - 5_000)
|
|
self.assertEqual(summary["anomalies"], [])
|
|
self.assertEqual(report["mismatches"], [])
|
|
|
|
|
|
class SimulatorVirtualCapTests(unittest.TestCase):
|
|
def test_simulator_rejects_actions_beyond_virtual_cap(self):
|
|
"""附件2 §4.5 兜底：虚拟时长达限后本地模拟器不再执行检测。"""
|
|
simulator = LocalSimulator([], robot_id="T001", config=CONFIG)
|
|
simulator.entered = True
|
|
simulator.virtual_time_s = CONFIG.field.max_virtual_duration_s
|
|
status, body = simulator.handle(
|
|
"/measure", {"arena_id": "default", "robot_id": "T001",
|
|
"request_id": "x1",
|
|
"position": {"x": 0.0, "y": 0.0}, "channel": 1})
|
|
self.assertEqual(status, 200)
|
|
self.assertFalse(body["accepted"])
|
|
self.assertEqual(body["virtual_time_s"], 0)
|
|
|
|
|
|
class AbortFinalizationTests(unittest.TestCase):
|
|
"""断连超过重试上限：仍须 /exit、产出未完成 summary，不得静默崩溃。"""
|
|
|
|
def test_transport_failure_aborts_with_summary_and_exit(self):
|
|
"""首个检测动作三次断连：run() 正常返回，/exit 成功，stop_reason 留档。"""
|
|
summary, _, _ = _run_scene(
|
|
random_omni_scene(seed=5, count=2),
|
|
fault_script=[None, "drop_before", "drop_before", "drop_before"])
|
|
self.assertFalse(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "transport_failure")
|
|
self.assertEqual(summary["exit_reason"], "user_exit",
|
|
"接口仍可用时必须尝试退出")
|
|
self.assertEqual(summary["abort_detail"]["type"], "TransportError")
|
|
self.assertLessEqual(summary["measures"] + summary["clears"],
|
|
summary["total_requests"])
|
|
self.assertEqual(len(summary["channel_status"]), 20,
|
|
"未完成状态须有逐频道快照")
|
|
|
|
def test_exit_failure_still_produces_summary(self):
|
|
"""检测与 /exit 均三次断连：产出 exit_unreachable 的未完成 summary。"""
|
|
summary, _, _ = _run_scene(
|
|
random_omni_scene(seed=7, count=2),
|
|
fault_script=[None,
|
|
"drop_before", "drop_before", "drop_before",
|
|
"drop_before", "drop_before", "drop_before"])
|
|
self.assertFalse(summary["completed"])
|
|
self.assertEqual(summary["stop_reason"], "transport_failure")
|
|
self.assertEqual(summary["exit_reason"], "exit_unreachable")
|
|
self.assertEqual(summary["exit_error"]["type"], "TransportError")
|
|
self.assertIn("channel_status", summary)
|
|
|
|
|
|
class CoverageLedgerIntegrationTests(unittest.TestCase):
|
|
def test_grid_matches_plan_derivation(self):
|
|
points = coverage_points(CONFIG)
|
|
self.assertEqual(len(points), 49)
|
|
self.assertEqual(min(p[0] for p in points), -2100.0)
|
|
self.assertEqual(max(p[1] for p in points), 2100.0)
|
|
# 单元内任意点到最近网格点 <= 700*sqrt(2) <= 1000 的覆盖保证
|
|
worst = 0.0
|
|
for gx in range(-20, 21):
|
|
for gy in range(-20, 21):
|
|
g = (gx * 90.0, gy * 90.0)
|
|
if (g[0] ** 2 + g[1] ** 2) ** 0.5 <= 1800.0:
|
|
nearest = min(math_dist(g, p) for p in points)
|
|
worst = max(worst, nearest)
|
|
self.assertLessEqual(worst, 700.0 * 2 ** 0.5 + 1e-9)
|
|
|
|
|
|
class SceneGenerationTests(unittest.TestCase):
|
|
"""场景生成 [题面]：默认 10~16 源、Ω 内均匀、频道互异、r_eff∈[1000,1500]。"""
|
|
|
|
def test_default_count_within_problem_range(self):
|
|
for seed in range(20):
|
|
sources = random_omni_scene(seed=seed)
|
|
self.assertTrue(10 <= len(sources) <= 16, f"seed={seed}")
|
|
|
|
def test_scene_respects_problem_constraints(self):
|
|
for seed in range(10):
|
|
sources = random_omni_scene(seed=seed)
|
|
channels = [s.channel for s in sources]
|
|
self.assertEqual(len(channels), len(set(channels)))
|
|
self.assertTrue(all(1 <= c <= 20 for c in channels))
|
|
for s in sources:
|
|
self.assertLessEqual((s.x ** 2 + s.y ** 2) ** 0.5, 1800.0)
|
|
self.assertGreaterEqual(s.r_eff_m, 1000.0)
|
|
self.assertLessEqual(s.r_eff_m, 1500.0)
|
|
|
|
|
|
def math_dist(a, b):
|
|
return ((a[0] - b[0]) ** 2 + (a[1] - b[1]) ** 2) ** 0.5
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|