diff --git a/diffusion_policy/env_runner/blockpush_lowdim_runner.py b/diffusion_policy/env_runner/blockpush_lowdim_runner.py
index b3f31bf..5193daf 100644
--- a/diffusion_policy/env_runner/blockpush_lowdim_runner.py
+++ b/diffusion_policy/env_runner/blockpush_lowdim_runner.py
@@ -235,7 +235,15 @@ class BlockPushLowdimRunner(BaseLowdimRunner):
         prefix_counts = collections.defaultdict(lambda : 0)
 
         log_data = dict()
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             this_rewards = all_rewards[i]
diff --git a/diffusion_policy/env_runner/kitchen_lowdim_runner.py b/diffusion_policy/env_runner/kitchen_lowdim_runner.py
index 417294b..39f5623 100644
--- a/diffusion_policy/env_runner/kitchen_lowdim_runner.py
+++ b/diffusion_policy/env_runner/kitchen_lowdim_runner.py
@@ -279,7 +279,15 @@ class KitchenLowdimRunner(BaseLowdimRunner):
         log_data = dict()
         prefix_total_reward_map = collections.defaultdict(list)
         prefix_n_completed_map = collections.defaultdict(list)
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             this_rewards = all_rewards[i]
diff --git a/diffusion_policy/env_runner/pusht_image_runner.py b/diffusion_policy/env_runner/pusht_image_runner.py
index 3c7fa09..f65c06a 100644
--- a/diffusion_policy/env_runner/pusht_image_runner.py
+++ b/diffusion_policy/env_runner/pusht_image_runner.py
@@ -221,7 +221,15 @@ class PushTImageRunner(BaseImageRunner):
         # log
         max_rewards = collections.defaultdict(list)
         log_data = dict()
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             max_reward = np.max(all_rewards[i])
diff --git a/diffusion_policy/env_runner/pusht_keypoints_runner.py b/diffusion_policy/env_runner/pusht_keypoints_runner.py
index b9c107c..a16bd58 100644
--- a/diffusion_policy/env_runner/pusht_keypoints_runner.py
+++ b/diffusion_policy/env_runner/pusht_keypoints_runner.py
@@ -243,7 +243,15 @@ class PushTKeypointsRunner(BaseLowdimRunner):
         # log
         max_rewards = collections.defaultdict(list)
         log_data = dict()
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             max_reward = np.max(all_rewards[i])
diff --git a/diffusion_policy/env_runner/robomimic_image_runner.py b/diffusion_policy/env_runner/robomimic_image_runner.py
index ade77ea..cbae74e 100644
--- a/diffusion_policy/env_runner/robomimic_image_runner.py
+++ b/diffusion_policy/env_runner/robomimic_image_runner.py
@@ -324,7 +324,15 @@ class RobomimicImageRunner(BaseImageRunner):
         # log
         max_rewards = collections.defaultdict(list)
         log_data = dict()
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             max_reward = np.max(all_rewards[i])
diff --git a/diffusion_policy/env_runner/robomimic_lowdim_runner.py b/diffusion_policy/env_runner/robomimic_lowdim_runner.py
index 7246943..f3ba642 100644
--- a/diffusion_policy/env_runner/robomimic_lowdim_runner.py
+++ b/diffusion_policy/env_runner/robomimic_lowdim_runner.py
@@ -317,7 +317,15 @@ class RobomimicLowdimRunner(BaseLowdimRunner):
         # log
         max_rewards = collections.defaultdict(list)
         log_data = dict()
-        for i in range(len(self.env_fns)):
+        # results reported in the paper are generated using the commented out line below
+        # which will only report and average metrics from first n_envs initial condition and seeds
+        # fortunately this won't invalidate our conclusion since
+        # 1. This bug only affects the variance of metrics, not their mean
+        # 2. All baseline methods are evaluated using the same code
+        # to completely reproduce reported numbers, uncomment this line:
+        # for i in range(len(self.env_fns)):
+        # and comment out this line
+        for i in range(n_inits):
             seed = self.env_seeds[i]
             prefix = self.env_prefixs[i]
             max_reward = np.max(all_rewards[i])