fix(rl): mirror gym_manipulator in actor

(cherry picked from commit d2a046dfc5)

src/lerobot/rl/actor.py

@@ -90,9 +90,9 @@ from lerobot.utils.utils import (
 )
 
 from .gym_manipulator import (
-    create_transition,
     make_processors,
     make_robot_env,
+    reset_and_build_transition,
     step_env_and_process_transition,
 )
 
@@ -265,13 +265,7 @@ def act_with_policy(
         dataset_stats=cfg.policy.dataset_stats,
     )
 
-    obs, info = online_env.reset()
+    transition = reset_and_build_transition(online_env, env_processor, action_processor)
-    env_processor.reset()
-    action_processor.reset()
-
-    # Process initial observation
-    transition = create_transition(observation=obs, info=info)
-    transition = env_processor(transition)
 
     # NOTE: For the moment we will solely handle the case of a single environment
     sum_reward_episode = 0
@@ -396,14 +390,7 @@ def act_with_policy(
             episode_intervention_steps = 0
             episode_total_steps = 0
 
-            # Reset environment and processors
+            transition = reset_and_build_transition(online_env, env_processor, action_processor)
-            obs, info = online_env.reset()
-            env_processor.reset()
-            action_processor.reset()
-
-            # Process initial observation
-            transition = create_transition(observation=obs, info=info)
-            transition = env_processor(transition)
 
         if cfg.env.fps is not None:
             dt_time = time.perf_counter() - start_time