Spaces:
Sleeping
Sleeping
File size: 17,110 Bytes
9b1756a | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 | """Pulse-backed OpenEnv environment implementation."""
from __future__ import annotations
import random
from uuid import uuid4
from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import EnvironmentMetadata, State
try:
from ..models import PulsePhysiologyAction, PulsePhysiologyObservation
from ..patient_state import PatientState
from ..runtime_effects import NoisyObservation, TimePressureMechanic
from ..tool_catalog import canonicalize_tool_name, coerce_boolean_argument
from .atls_judge import ATLSJudge
from .pathology_architect import PathologyArchitect, PathologyBlueprint
from .patient_monitor import PatientMonitorVisualization
from .pulse_engine_adapter import PulseEngineAdapter
from .reward_engine import RewardBreakdown, RewardEngine, RewardTracker
from .scenarios import DEFAULT_SCENARIO_ID, PatientProfile, ScenarioDefinition, get_scenario_definition
from .tools import PulseToolExecutor
except ImportError:
from models import PulsePhysiologyAction, PulsePhysiologyObservation
from patient_state import PatientState
from runtime_effects import NoisyObservation, TimePressureMechanic
from tool_catalog import canonicalize_tool_name, coerce_boolean_argument
from server.atls_judge import ATLSJudge
from server.pathology_architect import PathologyArchitect, PathologyBlueprint
from server.patient_monitor import PatientMonitorVisualization
from server.pulse_engine_adapter import PulseEngineAdapter
from server.reward_engine import RewardBreakdown, RewardEngine, RewardTracker
from server.scenarios import DEFAULT_SCENARIO_ID, PatientProfile, ScenarioDefinition, get_scenario_definition
from server.tools import PulseToolExecutor
class PulsePhysiologyEnvironment(Environment):
"""A Pulse-backed tool environment for trauma and resuscitation workflows."""
SUPPORTS_CONCURRENT_SESSIONS: bool = True
def __init__(
self,
*,
observation_noise_level: float = 0.0,
time_pressure_enabled: bool = False,
time_pressure_onset_s: float = 180.0,
time_pressure_escalation_per_minute: float = 0.15,
) -> None:
self._adapter = PulseEngineAdapter()
self._tool_executor = PulseToolExecutor(self._adapter)
self._reward_engine = RewardEngine()
self._monitor = PatientMonitorVisualization()
self._atls_judge = ATLSJudge()
self._pathology_architect = PathologyArchitect()
self._state = State(episode_id=str(uuid4()), step_count=0)
self._scenario: ScenarioDefinition = get_scenario_definition(DEFAULT_SCENARIO_ID)
self._selected_patient: PatientProfile | None = None
self._latest_patient_state: PatientState | None = None
self._reward_tracker: RewardTracker | None = None
self._last_reward_breakdown = RewardBreakdown()
self._active_blueprint: PathologyBlueprint | None = None
self._state_history: list[PatientState] = []
self._observed_state_history: list[PatientState] = []
self._observation_noise = NoisyObservation(observation_noise_level)
self._time_pressure = TimePressureMechanic(
enabled=time_pressure_enabled,
onset_s=time_pressure_onset_s,
escalation_per_minute=time_pressure_escalation_per_minute,
)
self._observation_rng = random.Random()
def reset(
self,
seed: int | None = None,
episode_id: str | None = None,
**kwargs: object,
) -> PulsePhysiologyObservation:
"""Reset the environment and initialize the requested Pulse scenario."""
self._configure_runtime_effects(kwargs)
try:
blueprint = self._resolve_generated_blueprint(kwargs)
except (KeyError, TypeError, ValueError) as exc:
raise ValueError(
"Invalid generated pathology inputs. Provide patient_id, injury_type, and severity."
) from exc
if blueprint is not None:
self._active_blueprint = blueprint
self._scenario = self._pathology_architect.to_scenario_definition(blueprint)
else:
self._active_blueprint = None
scenario_id = kwargs.get("scenario_id")
try:
self._scenario = get_scenario_definition(
str(scenario_id) if scenario_id is not None else DEFAULT_SCENARIO_ID
)
except KeyError as exc:
raise ValueError(str(exc)) from exc
self._state = State(episode_id=episode_id or str(uuid4()), step_count=0)
rng = random.Random(seed)
self._observation_rng = random.Random(rng.random())
self._selected_patient = self._scenario.choose_patient(rng)
patient_state = self._adapter.load_patient(
state_file=self._selected_patient.state_file,
scenario_id=self._scenario.scenario_id,
scenario_difficulty=self._scenario.difficulty,
patient_id=self._selected_patient.patient_id,
)
if self._scenario.setup is not None:
self._scenario.setup(self._adapter)
patient_state = self._adapter.get_full_state()
patient_state = self._apply_episode_rules(patient_state)
self._latest_patient_state = patient_state
self._reward_tracker = self._reward_engine.start_episode(self._scenario, patient_state)
self._last_reward_breakdown = RewardBreakdown(
reward_profile=self._scenario.reward_profile,
difficulty_multiplier=self._reward_engine.DIFFICULTY_MULTIPLIER[self._scenario.difficulty],
action_budget_remaining=self._reward_tracker.action_budget_remaining,
)
self._state_history = [patient_state]
self._observed_state_history = []
return self._build_observation(
patient_state,
reward=0.0,
tool_result=None,
error=None,
)
def step(
self,
action: PulsePhysiologyAction,
timeout_s: float | None = None,
**kwargs: object,
) -> PulsePhysiologyObservation:
"""Execute a named tool against the current Pulse scenario."""
del timeout_s, kwargs
self._state.step_count += 1
action = action.model_copy(
update={
"tool_name": canonicalize_tool_name(
action.tool_name,
allowed_tools=self._tool_executor.available_tools,
)
}
)
previous_state = self._latest_patient_state or self._apply_episode_rules(self._adapter.get_full_state())
execution = self._tool_executor.execute(action)
current_state = self._apply_episode_rules(execution.state)
if self._reward_tracker is None:
self._reward_tracker = self._reward_engine.start_episode(self._scenario, previous_state)
breakdown = self._reward_engine.score_step(
self._reward_tracker,
scenario=self._scenario,
before=previous_state,
after=current_state,
action=action,
success=execution.tool_result.success,
had_error=execution.error is not None,
time_pressure_multiplier=self._time_pressure.deterioration_multiplier(
sim_time_s=current_state.sim_time_s,
injury_severity=self._current_injury_severity(),
unstable=self._is_state_unstable(current_state),
),
)
reward = breakdown.total
self._latest_patient_state = current_state
self._last_reward_breakdown = breakdown
self._state_history.append(current_state)
return self._build_observation(
current_state,
reward=reward,
tool_result=execution.tool_result,
error=execution.error,
)
@property
def state(self) -> State:
"""Get the current environment state."""
return self._state
def get_metadata(self) -> EnvironmentMetadata:
description = (
"Pulse-backed trauma environment with engine-correct tools for airway, breathing, "
"circulation, diagnostics, and decision support."
)
return EnvironmentMetadata(
name="PulsePhysiologyEnvironment",
description=description,
version="0.4.0",
author="OpenAI Codex",
)
def close(self) -> None:
self._adapter.close()
def _apply_episode_rules(self, state: PatientState) -> PatientState:
done = state.done or state.sim_time_s >= self._scenario.max_time_s
alerts = list(state.active_alerts)
if state.sim_time_s >= self._scenario.max_time_s and "time_limit_reached" not in alerts:
alerts.append("time_limit_reached")
return state.model_copy(update={"done": done, "active_alerts": alerts})
def _build_observation(
self,
state: PatientState,
*,
reward: float,
tool_result,
error,
) -> PulsePhysiologyObservation:
observed_state, runtime_effect_metadata = self._observe_state(state)
observed_history = [*self._observed_state_history, observed_state]
self._observed_state_history = observed_history
metadata = {
"step_count": self._state.step_count,
"scenario_description": self._scenario.description,
"scenario_difficulty": self._scenario.difficulty,
"reward_profile": self._scenario.reward_profile,
"patient_pool_size": len(self._scenario.patient_pool),
"selected_state_file": self._selected_patient.state_file if self._selected_patient is not None else None,
"action_budget_remaining": self._reward_tracker.action_budget_remaining if self._reward_tracker is not None else None,
"reward_breakdown": self._last_reward_breakdown.as_metadata(),
"available_tools": self._tool_executor.available_tools,
"patient_monitor": self._monitor.build(
history=observed_history or [observed_state],
action_history=self._reward_tracker.action_history if self._reward_tracker is not None else [],
current_state=observed_state,
).as_dict(),
"atls_judge": self._atls_judge.evaluate(
state=state,
action_history=self._reward_tracker.action_history if self._reward_tracker is not None else [],
reward_profile=self._scenario.reward_profile,
state_history=self._state_history,
).as_dict(),
"pathology_blueprint": self._active_blueprint.as_dict() if self._active_blueprint is not None else None,
**runtime_effect_metadata,
}
return PulsePhysiologyObservation.from_patient_state(
observed_state,
reward=reward,
available_tools=self._tool_executor.available_tools,
tool_result=tool_result,
error=error,
metadata=metadata,
)
def _resolve_generated_blueprint(self, kwargs: dict[str, object]) -> PathologyBlueprint | None:
has_scenario_id = kwargs.get("scenario_id") is not None
raw_blueprint = kwargs.get("pathology_blueprint")
base_keys = {"patient_id", "severity"}
selector_keys = {"injury_type", "injury_types"}
provided_base_keys = {key for key in base_keys if kwargs.get(key) is not None}
provided_selector_keys = {key for key in selector_keys if kwargs.get(key) is not None}
provided_authoring_keys = provided_base_keys | provided_selector_keys
if has_scenario_id and (raw_blueprint is not None or provided_authoring_keys):
raise ValueError(
"Pass either scenario_id or generated-case authoring inputs, not both."
)
if raw_blueprint is not None and provided_authoring_keys:
raise ValueError(
"Pass either pathology_blueprint or patient_id/injury_type(s)/severity, not both."
)
if isinstance(raw_blueprint, dict):
raw_base_keys = {key for key in base_keys if raw_blueprint.get(key) is not None}
raw_selector_keys = {key for key in selector_keys if raw_blueprint.get(key) is not None}
missing_base_keys = base_keys - raw_base_keys
if missing_base_keys:
missing = ", ".join(sorted(missing_base_keys))
raise ValueError(
f"pathology_blueprint is missing required keys: {missing}"
)
if not raw_selector_keys:
raise ValueError(
"pathology_blueprint must include injury_type or injury_types."
)
return self._pathology_architect.build_blueprint(
patient_id=str(raw_blueprint["patient_id"]),
injury_type=(
str(raw_blueprint["injury_type"])
if raw_blueprint.get("injury_type") is not None and raw_blueprint.get("injury_types") is None
else None
),
injury_types=raw_blueprint.get("injury_types"),
severity=float(raw_blueprint["severity"]),
)
if provided_authoring_keys and not provided_base_keys:
raise ValueError(
"Generated-case reset requires patient_id and severity together with injury_type or injury_types."
)
if provided_selector_keys and len(provided_selector_keys) != 1:
raise ValueError(
"Generated-case reset accepts exactly one of injury_type or injury_types."
)
if provided_base_keys and provided_base_keys != base_keys:
missing = ", ".join(sorted(base_keys - provided_base_keys))
raise ValueError(
f"Generated-case reset requires patient_id and severity together. Missing: {missing}"
)
if provided_base_keys == base_keys and not provided_selector_keys:
raise ValueError(
"Generated-case reset requires exactly one of injury_type or injury_types."
)
if provided_base_keys == base_keys and len(provided_selector_keys) == 1:
return self._pathology_architect.build_blueprint(
patient_id=str(kwargs["patient_id"]),
injury_type=str(kwargs["injury_type"]) if kwargs.get("injury_type") is not None else None,
injury_types=kwargs.get("injury_types"),
severity=float(kwargs["severity"]),
)
return None
def _configure_runtime_effects(self, kwargs: dict[str, object]) -> None:
observation_noise_level = float(kwargs.get("observation_noise_level", self._observation_noise.config.noise_level))
raw_time_pressure_enabled = kwargs.get("time_pressure_enabled", self._time_pressure.config.enabled)
if isinstance(raw_time_pressure_enabled, str):
time_pressure_enabled = coerce_boolean_argument(raw_time_pressure_enabled)
else:
time_pressure_enabled = bool(raw_time_pressure_enabled)
time_pressure_onset_s = float(kwargs.get("time_pressure_onset_s", self._time_pressure.config.onset_s))
time_pressure_escalation_per_minute = float(
kwargs.get(
"time_pressure_escalation_per_minute",
self._time_pressure.config.escalation_per_minute,
)
)
self._observation_noise = NoisyObservation(observation_noise_level)
self._time_pressure = TimePressureMechanic(
enabled=time_pressure_enabled,
onset_s=time_pressure_onset_s,
escalation_per_minute=time_pressure_escalation_per_minute,
)
def _observe_state(self, state: PatientState) -> tuple[PatientState, dict[str, object]]:
observed_state, noise_metadata = self._observation_noise.apply(
state,
rng=self._observation_rng,
)
time_pressure_metadata = self._time_pressure.as_metadata(
sim_time_s=state.sim_time_s,
injury_severity=self._current_injury_severity(),
unstable=self._is_state_unstable(state),
)
return observed_state, {
"observation_noise": noise_metadata,
"time_pressure": time_pressure_metadata,
}
def _current_injury_severity(self) -> float:
if self._active_blueprint is not None:
return float(self._active_blueprint.severity)
return {"easy": 0.35, "medium": 0.6, "hard": 0.85}[self._scenario.difficulty]
@staticmethod
def _is_state_unstable(state: PatientState) -> bool:
systolic = state.systolic_bp_mmhg if state.systolic_bp_mmhg is not None else 120.0
spo2 = state.spo2 if state.spo2 is not None else 1.0
return bool(state.active_alerts) or systolic < 95.0 or spo2 < 0.92 or state.mental_status != "alert"
|