fix: remove duplicate reward episode metrics

src/unilab/envs/manager_based_rl_env.py

@@ -629,8 +629,8 @@ def reset(
             if self._autoreset_reset_active:
                 # Autoreset runs at the tail of step(): keep this step's
                 # per-step log entries (reward/* etc., computed pre-reset) and
-                # layer the reset extras (Episode_Reward/* etc.) on top, so
-                # consumers still see the transition's reward breakdown.
+                # layer manager reset extras on top, so consumers still see the
+                # transition's metrics.
                 step_log = self._state.info.get("log")
                 if step_log:
                     log = {**step_log, **log}

src/unilab/managers/reward_manager.py

@@ -39,8 +39,6 @@ class RewardManager(ManagerBase):
 
       When ``scale_by_dt=True`` (default):
         - ``reward_buf`` (returned by ``compute()``) = raw_value * weight * dt
-        - ``_episode_sums`` (cumulative rewards) are scaled by dt
-        - ``Episode_Reward/*`` logged metrics are scaled by dt
 
       When ``scale_by_dt=False``:
         - ``reward_buf`` = raw_value * weight (no dt scaling)
@@ -71,9 +69,6 @@ def __init__(
 
         self.cfg = deepcopy(cfg)
         super().__init__(env=env)
-        self._episode_sums = dict()
-        for term_name in self._term_names:
-            self._episode_sums[term_name] = np.zeros(self.num_envs, dtype=np.float32)
         self._reward_buf = np.zeros(self.num_envs, dtype=np.float32)
         self._step_reward = np.zeros((self.num_envs, len(self._term_names)), dtype=np.float32)
         # Scratch for the weighted term value, reused across terms to avoid a
@@ -107,14 +102,9 @@ def active_terms(self) -> list[str]:
     def reset(self, env_ids: np.ndarray | slice | None = None) -> dict[str, float]:
         if env_ids is None:
             env_ids = slice(None)
-        extras = {}
-        for key in self._episode_sums.keys():
-            episodic_sum_avg = float(np.mean(self._episode_sums[key][env_ids]))
-            extras["Episode_Reward/" + key] = episodic_sum_avg / self._env.max_episode_length_s
-            self._episode_sums[key][env_ids] = 0.0
         for term_cfg in self._class_term_cfgs:
             term_cfg.func.reset(env_ids=env_ids)
-        return extras
+        return {}
 
     def compute(self, dt: float) -> np.ndarray:
         if not np.isfinite(dt) or (self._scale_by_dt and dt <= 0.0):
@@ -142,7 +132,6 @@ def compute(self, dt: float) -> np.ndarray:
             np.multiply(value, term_cfg.weight, out=scratch)
             scratch *= scale
             self._reward_buf += scratch
-            self._episode_sums[name] += scratch
             np.divide(scratch, scale, out=self._step_reward[:, term_idx])
         return self._reward_buf
 

tests/managers/test_core_managers.py

@@ -171,8 +171,11 @@ def test_reward_dt_scaling_reset_and_config_immutability(fake_env: FakeEnv) -> N
     manager = RewardManager(cfg, fake_env)
     np.testing.assert_allclose(manager.compute(dt=0.25), fake_env.value * 0.5)
     assert manager.get_active_iterable_terms(2) == [("stateful", [4.0])]
-    extras = manager.reset(np.array([1, 2]))
-    assert extras["Episode_Reward/stateful"] == pytest.approx(0.375)
+    reset_ids = np.array([1, 2])
+    extras = manager.reset(reset_ids)
+    assert extras == {}
+    assert not any(key.startswith("Episode_Reward/") for key in extras)
+    assert manager.get_term_cfg("stateful").func.reset_ids is reset_ids
     assert cfg["stateful"].func is StatefulReward
     assert isinstance(manager.get_term_cfg("stateful").func, StatefulReward)