File size: 3,230 Bytes
6aa6c12 | 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 | """Native eval_policy post-action ordering, isolated for independent regression.
All actors/poses/metrics here are private server state. This wrapper never
creates an action, repeats a control, or calls an expert.
"""
from __future__ import annotations
class NativeOutcome:
def __init__(self, env, config, tracker_class):
self.env = env
self.success = self.finished = self.has_lifted = False
self.fail_reason = None
self.check_z_threshold = self.target_actor = self.start_z = None
self.stop_on_contact = True
dynamic = getattr(env, "get_dynamic_motion_config", lambda: None)()
if dynamic and "check_z_threshold" in dynamic:
self.check_z_threshold = dynamic["check_z_threshold"]
self.target_actor = dynamic["check_z_actor"] if "check_z_actor" in dynamic else dynamic["target_actor"]
self.start_z = self.target_actor.get_pose().p[2]
if dynamic and "stop_on_contact" in dynamic:
self.stop_on_contact = dynamic["stop_on_contact"]
self.tracker = tracker_class(env, config)
self.tracker.on_episode_start()
env._metrics_tracker = self.tracker
def after_action(self):
"""Exactly eval_policy's post-eval_func checks, in native order."""
env = self.env
if self.check_z_threshold is not None and not self.has_lifted:
self.has_lifted = self.target_actor.get_pose().p[2] - self.start_z > self.check_z_threshold
if env.eval_success:
self.success = self.check_z_threshold is None or self.has_lifted
self.fail_reason = None if self.success else "object_not_lifted"
self.finished = True
return
if env.use_dynamic:
if env.check_gripper_contact_dynamic_object() and self.stop_on_contact:
env.stop_dynamic_object_motion()
if not getattr(env, "_dynamic_object_stopped", False) and not env.check_dynamic_object_boundary():
env.eval_fail = True
self.fail_reason = "out_of_bounds"
self.tracker.record_out_of_bounds()
self.finished = True
return
if env.eval_fail:
self.fail_reason = "eval_fail"
self.finished = True
return
self.finished = env.take_action_cnt >= env.step_lim
def metrics(self, seed):
value = self.tracker.get_episode_metrics(self.success, self.fail_reason, seed=seed)
return {"success": bool(value.success), "native_steps": int(value.steps_taken),
"metrics_tracker_steps": int(value.steps_taken),
"native_horizon": int(value.max_steps), "route_completion": float(value.route_completion),
"manipulation_score": float(value.manipulation_score),
"total_penalty_factor": float(value.total_penalty_factor),
"fail_reason": value.fail_reason, "native_seed": seed,
"penalty_events": [{"event_type": p.event_type, "timestep": int(p.timestep),
"penalty_factor": float(p.penalty_factor), "details": p.details}
for p in value.penalty_events]}
|