[RLlib] Minimal ParamNoise PR. (#7772)

rllib/evaluation/sample_batch_builder.py

@@ -151,6 +151,9 @@ class MultiAgentSampleBatchBuilder:
                     "from a single trajectory.", pre_batch)
             post_batches[agent_id] = policy.postprocess_trajectory(
                 pre_batch, other_batches, episode)
+            # Call the Policy's Exploration's postprocess method.
+            policy.exploration.postprocess_trajectory(
+                policy, post_batches[agent_id], getattr(policy, "_sess", None))
 
         if log_once("after_post"):
             logger.info(