diff --git a/diffusion_policy/config/train_pusht_pretrained_dinov2_base.yaml b/diffusion_policy/config/train_pusht_pretrained_dinov2_base.yaml
index 32fe19116..26bd9c72e 100644
--- a/diffusion_policy/config/train_pusht_pretrained_dinov2_base.yaml
+++ b/diffusion_policy/config/train_pusht_pretrained_dinov2_base.yaml
@@ -2,7 +2,7 @@ defaults:
   - _self_
   - task: pusht_image
 
-name: pusht_pretrained_dinov2_base
+name: pusht_pretrained_dinov2_base_preprocessing
 _target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
 
 task_name: ${task.name}
@@ -45,6 +45,7 @@ policy:
     use_group_norm: False
     share_rgb_model: True
     imagenet_norm: False
+    rgb_model_name: facebook/dinov2-base
 
   horizon: ${horizon}
   n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
diff --git a/diffusion_policy/config/train_pusht_pretrained_dinov2_large.yaml b/diffusion_policy/config/train_pusht_pretrained_dinov2_large.yaml
index fb006d701..b424444ae 100644
--- a/diffusion_policy/config/train_pusht_pretrained_dinov2_large.yaml
+++ b/diffusion_policy/config/train_pusht_pretrained_dinov2_large.yaml
@@ -2,7 +2,7 @@ defaults:
   - _self_
   - task: pusht_image
 
-name: pusht_pretrained_dinov2_large
+name: pusht_pretrained_dinov2_large_preprocessing
 _target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
 
 task_name: ${task.name}
@@ -45,6 +45,7 @@ policy:
     use_group_norm: False
     share_rgb_model: True
     imagenet_norm: False
+    rgb_model_name: facebook/dinov2-large
 
   horizon: ${horizon}
   n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
diff --git a/diffusion_policy/config/train_real_pusht.yaml b/diffusion_policy/config/train_real_pusht.yaml
new file mode 100644
index 000000000..226be6274
--- /dev/null
+++ b/diffusion_policy/config/train_real_pusht.yaml
@@ -0,0 +1,152 @@
+defaults:
+  - _self_
+  - task: real_pusht_image
+
+name: real_pusht_end2end
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_resnet
+      name: resnet18
+      weights: null
+    resize_shape: [224,224]
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: True
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 100
+  num_epochs: 600
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: False
+  # training loop control
+  # in epochs
+  rollout_every: 50
+  checkpoint_every: 50
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_real_pusht_pretrained_dinov2_base.yaml b/diffusion_policy/config/train_real_pusht_pretrained_dinov2_base.yaml
new file mode 100644
index 000000000..6f1ffdd3c
--- /dev/null
+++ b/diffusion_policy/config/train_real_pusht_pretrained_dinov2_base.yaml
@@ -0,0 +1,149 @@
+defaults:
+  - _self_
+  - task: real_pusht_image
+
+name: real_pusht_dinov2_base_preprocessing
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+  shape_meta: ${shape_meta}
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_dinov2
+      name: facebook/dinov2-base
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: False
+    rgb_model_name: facebook/dinov2-base
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 100
+  num_epochs: 600
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 50
+  checkpoint_every: 50
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_real_pusht_pretrained_dinov2_large.yaml b/diffusion_policy/config/train_real_pusht_pretrained_dinov2_large.yaml
new file mode 100644
index 000000000..7c67c8db0
--- /dev/null
+++ b/diffusion_policy/config/train_real_pusht_pretrained_dinov2_large.yaml
@@ -0,0 +1,149 @@
+defaults:
+  - _self_
+  - task: real_pusht_image
+
+name: real_pusht_dinov2_large_preprocessing
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+  shape_meta: ${shape_meta}
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_dinov2
+      name: facebook/dinov2-large
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: False
+    rgb_model_name: facebook/dinov2-large
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 100
+  num_epochs: 600
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 50
+  checkpoint_every: 50
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_real_pusht_pretrained_imagenet.yaml b/diffusion_policy/config/train_real_pusht_pretrained_imagenet.yaml
new file mode 100644
index 000000000..ec2b59518
--- /dev/null
+++ b/diffusion_policy/config/train_real_pusht_pretrained_imagenet.yaml
@@ -0,0 +1,152 @@
+defaults:
+  - _self_
+  - task: real_pusht_image
+
+name: real_pusht_pretrained_imagenet
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_resnet
+      name: resnet18
+      weights: IMAGENET1K_V1 # or r3m
+    # resize_shape: [224,224]
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: True
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 100
+  num_epochs: 600
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 50
+  checkpoint_every: 50
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_real_pusht_pretrained_r3m.yaml b/diffusion_policy/config/train_real_pusht_pretrained_r3m.yaml
new file mode 100644
index 000000000..61b25a21f
--- /dev/null
+++ b/diffusion_policy/config/train_real_pusht_pretrained_r3m.yaml
@@ -0,0 +1,152 @@
+defaults:
+  - _self_
+  - task: real_pusht_image
+
+name: real_pusht_pretrained_r3m
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_resnet
+      name: resnet18
+      weights: r3m # or r3m
+    # resize_shape: [224,224]
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: True
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 100
+  num_epochs: 600
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 50
+  checkpoint_every: 50
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_robomimic.yaml b/diffusion_policy/config/train_robomimic.yaml
new file mode 100644
index 000000000..4813bf16a
--- /dev/null
+++ b/diffusion_policy/config/train_robomimic.yaml
@@ -0,0 +1,152 @@
+defaults:
+  - _self_
+  - task: can_image_abs
+
+name: can_ph_end2end
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_resnet
+      name: resnet18
+      weights: null
+    # resize_shape: [240, 320]
+    crop_shape: [76, 76] # ch, cw 240x320 90%
+    random_crop: True
+    use_group_norm: True
+    share_rgb_model: False
+    imagenet_norm: True
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 66
+  num_epochs: 400
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: False
+  # training loop control
+  # in epochs
+  rollout_every: 20
+  checkpoint_every: 20
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_robomimic_dinov2.yaml b/diffusion_policy/config/train_robomimic_dinov2.yaml
new file mode 100644
index 000000000..e66efc80e
--- /dev/null
+++ b/diffusion_policy/config/train_robomimic_dinov2.yaml
@@ -0,0 +1,151 @@
+defaults:
+  - _self_
+  - task: can_image_abs
+
+name: can_ph_dino
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_dinov2
+      name: facebook/dinov2-base
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: False
+    rgb_model_name: facebook/dinov2-base
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 66
+  num_epochs: 400
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 20
+  checkpoint_every: 20
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/config/train_robomimic_r3m.yaml b/diffusion_policy/config/train_robomimic_r3m.yaml
new file mode 100644
index 000000000..779154d32
--- /dev/null
+++ b/diffusion_policy/config/train_robomimic_r3m.yaml
@@ -0,0 +1,152 @@
+defaults:
+  - _self_
+  - task: can_image_abs
+
+name: can_ph_r3m
+_target_: diffusion_policy.workspace.train_diffusion_unet_image_workspace.TrainDiffusionUnetImageWorkspace
+
+task_name: ${task.name}
+shape_meta: ${task.shape_meta}
+exp_name: "default"
+
+horizon: 16
+n_obs_steps: 2
+n_action_steps: 8
+n_latency_steps: 0
+dataset_obs_steps: ${n_obs_steps}
+past_action_visible: False
+keypoint_visible_rate: 1.0
+obs_as_global_cond: True
+
+policy:
+  _target_: diffusion_policy.policy.diffusion_unet_image_policy.DiffusionUnetImagePolicy
+
+  shape_meta: ${shape_meta}
+  
+  noise_scheduler:
+    _target_: diffusers.schedulers.scheduling_ddim.DDIMScheduler
+    num_train_timesteps: 100
+    beta_start: 0.0001
+    beta_end: 0.02
+    # beta_schedule is important
+    # this is the best we found
+    beta_schedule: squaredcos_cap_v2
+    clip_sample: True
+    set_alpha_to_one: True
+    steps_offset: 0
+    prediction_type: epsilon # or sample
+
+  obs_encoder:
+    _target_: diffusion_policy.model.vision.multi_image_obs_encoder.MultiImageObsEncoder
+    shape_meta: ${shape_meta}
+    rgb_model:
+      _target_: diffusion_policy.model.vision.model_getter.get_resnet
+      name: resnet18
+      weights: r3m # or r3m
+    # resize_shape: [224,224]
+    crop_shape: null
+    random_crop: False
+    use_group_norm: False
+    share_rgb_model: True
+    imagenet_norm: True
+
+  horizon: ${horizon}
+  n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
+  n_obs_steps: ${n_obs_steps}
+  num_inference_steps: 100
+  obs_as_global_cond: ${obs_as_global_cond}
+  # crop_shape: null
+  diffusion_step_embed_dim: 128
+  down_dims: [512, 1024, 2048]
+  kernel_size: 5
+  n_groups: 8
+  cond_predict_scale: True
+
+  # scheduler.step params
+  # predict_epsilon: True
+
+ema:
+  _target_: diffusion_policy.model.diffusion.ema_model.EMAModel
+  update_after_step: 0
+  inv_gamma: 1.0
+  power: 0.75
+  min_value: 0.0
+  max_value: 0.9999
+
+dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: True
+  pin_memory: True
+  persistent_workers: True
+
+val_dataloader:
+  batch_size: 64
+  num_workers: 8
+  shuffle: False
+  pin_memory: True
+  persistent_workers: True
+
+optimizer:
+  _target_: torch.optim.AdamW
+  lr: 1.0e-4
+  betas: [0.95, 0.999]
+  eps: 1.0e-8
+  weight_decay: 1.0e-6
+
+training:
+  device: "cuda:0"
+  seed: 42
+  debug: False
+  resume: True
+  # optimization
+  lr_scheduler: cosine
+  lr_warmup_steps: 66
+  num_epochs: 400
+  gradient_accumulate_every: 1
+  # EMA destroys performance when used with BatchNorm
+  # replace BatchNorm with GroupNorm.
+  use_ema: True
+  freeze_encoder: True
+  # training loop control
+  # in epochs
+  rollout_every: 20
+  checkpoint_every: 20
+  val_every: 1
+  sample_every: 5
+  # steps per epoch
+  max_train_steps: null
+  max_val_steps: null
+  # misc
+  tqdm_interval_sec: 1.0
+
+logging:
+  project: diffusion_policy_debug
+  resume: True
+  mode: online
+  name: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+  tags: ["${name}", "${task_name}", "${exp_name}"]
+  id: null
+  group: null
+
+checkpoint:
+  topk:
+    monitor_key: train_loss
+    mode: min
+    k: 5
+    format_str: 'epoch={epoch:04d}-train_loss={train_loss:.3f}.ckpt'
+  save_last_ckpt: True
+  save_last_snapshot: False
+
+multi_run:
+  run_dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  wandb_name_base: ${now:%Y.%m.%d-%H.%M.%S}_${name}_${task_name}
+
+hydra:
+  job:
+    override_dirname: ${name}
+  run:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+  sweep:
+    dir: data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}
+    subdir: ${hydra.job.num}
diff --git a/diffusion_policy/model/vision/multi_image_obs_encoder.py b/diffusion_policy/model/vision/multi_image_obs_encoder.py
index ccd76b07d..fc9fda5e2 100644
--- a/diffusion_policy/model/vision/multi_image_obs_encoder.py
+++ b/diffusion_policy/model/vision/multi_image_obs_encoder.py
@@ -5,6 +5,7 @@
 import torch.nn as nn
 import torchvision
 import transformers
+from transformers import AutoImageProcessor
 
 from diffusion_policy.common.pytorch_util import dict_apply, replace_submodules
 from diffusion_policy.model.common.module_attr_mixin import ModuleAttrMixin
@@ -24,7 +25,8 @@ def __init__(self,
             share_rgb_model: bool=False,
             # renormalize rgb input with imagenet normalization
             # assuming input in [0,1]
-            imagenet_norm: bool=False
+            imagenet_norm: bool=False,
+            rgb_model_name: str=None,
         ):
         """
         Assumes rgb input: B,C,H,W
@@ -126,6 +128,13 @@ def __init__(self,
         self.rgb_keys = rgb_keys
         self.low_dim_keys = low_dim_keys
         self.key_shape_map = key_shape_map
+        self.rgb_model_name = rgb_model_name
+        self.predefined_image_processor = None
+        if (
+            self.rgb_model_name == "facebook/dinov2-base"
+            or self.rgb_model_name == "facebook/dinov2-large"
+        ):
+            self.predefined_image_processor = AutoImageProcessor.from_pretrained(self.rgb_model_name)
 
     def forward(self, obs_dict):
         batch_size = None
@@ -146,8 +155,18 @@ def forward(self, obs_dict):
             # (N*B,C,H,W)
             imgs = torch.cat(imgs, dim=0)
             # (N*B,D)
-            if isinstance(self.key_model_map['rgb'], transformers.models.dinov2.modeling_dinov2.Dinov2Model):
-                feature = self.key_model_map["rgb"](imgs)[0].mean(dim=1)
+            if (
+                self.rgb_model_name == "facebook/dinov2-base"
+                or self.rgb_model_name == "facebook/dinov2-large"
+            ):
+                imgs = (imgs * 255).type(torch.uint8)
+                inputs = self.predefined_image_processor(
+                    images=imgs, 
+                    return_tensors="pt",
+                )
+                inputs['pixel_values'] = inputs['pixel_values'].to(self.device)
+                outputs = self.key_model_map["rgb"](**inputs)
+                feature = outputs[0].mean(dim=1)
             else:
                 feature = self.key_model_map["rgb"](imgs)
             # (N,B,D)
diff --git a/scripts/fix_mujoco.sh b/scripts/fix_mujoco.sh
new file mode 100755
index 000000000..d4b0c89ad
--- /dev/null
+++ b/scripts/fix_mujoco.sh
@@ -0,0 +1,43 @@
+#!/bin/bash
+set -e
+
+echo "Activating environment..."
+eval "$(conda shell.bash hook)"
+
+mamba activate robodiff
+mamba install -c conda-forge glew
+mamba install -c conda-forge mesalib
+mamba install -c anaconda mesa-libgl-cos6-x86_64
+mamba install -c menpo glfw3
+
+conda env config vars set MUJOCO_GL=egl PYOPENGL_PLATFORM=egl
+conda deactivate && conda activate robodiff
+
+pip install patchelf
+
+echo "Adding environment variable to bashrc"
+#!/usr/bin/env bash
+
+# Lines to add
+LINES=(
+'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/portegak/.mujoco/mujoco210/bin'
+'#export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia'
+'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/miniforge3/envs/robodiff/include'
+)
+
+BASHRC="$HOME/.bashrc"
+
+# Check if each line already exists in .bashrc, if not, append it
+for LINE in "${LINES[@]}"; do
+    # Escape special characters for grep
+    ESCAPED_LINE=$(printf '%s\n' "$LINE" | sed 's/[][\.*^$(){}?+|]/\\&/g')
+
+    if ! grep -qxF "$LINE" "$BASHRC" 2>/dev/null; then
+        echo "$LINE" >> "$BASHRC"
+        echo "Added: $LINE"
+    else
+        echo "Already present: $LINE"
+    fi
+done
+
+echo "Done. To apply changes, run: source $BASHRC"
diff --git a/scripts/run_eval.sh b/scripts/run_eval.sh
new file mode 100755
index 000000000..7cd23c63b
--- /dev/null
+++ b/scripts/run_eval.sh
@@ -0,0 +1,3 @@
+
+CUDA_VISIBLE_DEVICES=1
+python eval.py --checkpoint data/outputs/2024.12.10/14.11.59_real_pusht_end2end_real_image/checkpoints/latest.ckpt -o data/pusht_real_eval_output
diff --git a/scripts/train_dino.sh b/scripts/train_dino.sh
new file mode 100755
index 000000000..763f168ce
--- /dev/null
+++ b/scripts/train_dino.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+set -e
+CONFIG_FOLDER="diffusion_policy/config"
+
+echo  "Running PushT with DinoBase"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_pusht_pretrained_dinov2_base.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo  "Running PushT with DinoLarge"
+python train.py --config-dir=$CONFIG_FOLDER  --config-name=train_pusht_pretrained_dinov2_large.yaml training.seed=42 training.device=cuda:1 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
\ No newline at end of file
diff --git a/scripts/train_real_pusht _gpu1.sh b/scripts/train_real_pusht _gpu1.sh
new file mode 100755
index 000000000..4ca03078d
--- /dev/null
+++ b/scripts/train_real_pusht _gpu1.sh	
@@ -0,0 +1,16 @@
+#!/bin/bash
+set -e
+CONFIG_FOLDER="diffusion_policy/config"
+
+echo  "Running RealPushT with End2End"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with DinoV2"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_dinov2_base.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with R3M"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_r3m.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with ImageNet"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_imagenet.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
diff --git a/scripts/train_real_pusht.sh b/scripts/train_real_pusht.sh
new file mode 100755
index 000000000..4ca03078d
--- /dev/null
+++ b/scripts/train_real_pusht.sh
@@ -0,0 +1,16 @@
+#!/bin/bash
+set -e
+CONFIG_FOLDER="diffusion_policy/config"
+
+echo  "Running RealPushT with End2End"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with DinoV2"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_dinov2_base.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with R3M"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_r3m.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo "Running RealPushT with ImageNet"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_real_pusht_pretrained_imagenet.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
diff --git a/scripts/train_robomimic.sh b/scripts/train_robomimic.sh
new file mode 100755
index 000000000..b33fdd5a2
--- /dev/null
+++ b/scripts/train_robomimic.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+set -e
+CONFIG_FOLDER="diffusion_policy/config"
+
+echo  "Running Can lift with DinoBase"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_robomimic_dinov2.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo  "Running Can lift with End2End"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_robomimic.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+
+echo  "Running Can lift with R3M"
+python train.py --config-dir=$CONFIG_FOLDER --config-name=train_robomimic_r3m.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'
+