COMAA runs

config/ppo/PushBlock.yaml

         gamma: 0.99
         strength: 1.0
     keep_checkpoints: 5
-    max_steps: 2000000
-    time_horizon: 64
-    summary_freq: 60000
+    max_steps: 20000000
+    time_horizon: 1000
+    summary_freq: 10000
     threaded: true

ml-agents/mlagents/trainers/trajectory.py

                 for teammate_status in next_exp.teammate_status:
                     teammate_cont_next_actions.append(teammate_status.action.continuous)
                     teammate_disc_next_actions.append(teammate_status.action.discrete)
+            else:
+                for teammate_status in exp.teammate_status:
+                    teammate_cont_next_actions.append(teammate_status.action.continuous)
+                    teammate_disc_next_actions.append(teammate_status.action.discrete)
+
             agent_buffer_trajectory["team_next_continuous_action"].append(
                 teammate_cont_next_actions
             )