[Attempted fix]

python/unitytrainers/ppo/trainer.py

         else:
             return run_out[self.model.output], None, None, run_out
 
-    def generate_intrinsic_rewards(self, curr_info, next_info):
+    def generate_intrinsic_rewards(self, next_info):
-        :param curr_info: Current BrainInfo.
-            feed_dict = {self.model.batch_size: len(curr_info.vector_observations), self.model.sequence_length: 1}
+            agent_index_to_ignore = []
+            for agent_index, agent_id in enumerate(next_info.agents):
+                if self.training_buffer[agent_id].last_brain_info is None:
+                    agent_index_to_ignore.append(agent_index)
+            feed_dict = {self.model.batch_size: len(next_info.vector_observations), self.model.sequence_length: 1}
-                for i in range(len(curr_info.visual_observations)):
-                    feed_dict[self.model.visual_in[i]] = curr_info.visual_observations[i]
+                for i in range(len(next_info.visual_observations)):
+                    tmp = []
+                    for agent_id in next_info.agents:
+                        agent_brain_info = self.training_buffer[agent_id].last_brain_info
+                        if agent_brain_info is None:
+                            agent_brain_info = next_info
+                        agent_obs = agent_brain_info.visual_observations[i][agent_brain_info.agents.index(agent_id)]
+                        tmp += [agent_obs]
+                    feed_dict[self.model.visual_in[i]] = np.array(tmp)
-                feed_dict[self.model.vector_in] = curr_info.vector_observations
+                tmp = []
+                for agent_id in next_info.agents:
+                    agent_brain_info = self.training_buffer[agent_id].last_brain_info
+                    if agent_brain_info is None:
+                        agent_brain_info = next_info
+                    agent_obs = agent_brain_info.vector_observations[agent_brain_info.agents.index(agent_id)]
+                    tmp += [agent_obs]
+                feed_dict[self.model.vector_in] = np.array(tmp)
-                if curr_info.memories.shape[1] == 0:
-                    curr_info.memories = np.zeros((len(curr_info.agents), self.m_size))
-                feed_dict[self.model.memory_in] = curr_info.memories
+                tmp = []
+                for agent_id in next_info.agents:
+                    agent_brain_info = self.training_buffer[agent_id].last_brain_info
+                    if agent_brain_info is None:
+                        agent_brain_info = next_info
+                    if agent_brain_info.memories.shape[1] == 0:
+                        agent_obs = np.zeros(self.m_size)
+                    else:
+                        agent_obs = agent_brain_info.memories[agent_brain_info.agents.index(agent_id)]
+                    tmp += [agent_obs]
+                feed_dict[self.model.memory_in] = np.array(tmp)
+            for index in agent_index_to_ignore:
+                intrinsic_rewards[index] = 0
             return intrinsic_rewards
         else:
             return None
         curr_info = curr_all_info[self.brain_name]
         next_info = next_all_info[self.brain_name]
 
-        intrinsic_rewards = self.generate_intrinsic_rewards(curr_info, next_info)
+        # intrinsic_rewards = self.generate_intrinsic_rewards(curr_info, next_info)
+
+        intrinsic_rewards = self.generate_intrinsic_rewards(next_info)
 
         for agent_id in next_info.agents:
             stored_info = self.training_buffer[agent_id].last_brain_info