Upload ITPS Maze2D pretrained checkpoint

README.md

@@ -0,0 +1,56 @@
+---
+library_name: lerobot
+license: mit
+tags:
+  - lerobot
+  - act
+  - robotics
+  - maze2d
+  - itps
+  - pytorch_model_hub_mixin
+pipeline_tag: robotics
+---
+
+# ITPS Maze2D — Action Chunking Transformer (ACT)
+
+Pre-trained Action Chunking Transformer checkpoint used in
+**Inference-Time Policy Steering through Human Interactions**
+([paper](https://huggingface.co/papers/2411.16627), [project page](https://yanweiw.github.io/itps/), [code](https://github.com/yanweiw/itps)).
+
+The model was trained on the [D4RL Maze2D](https://github.com/Farama-Foundation/D4RL)
+dataset and is intended to be loaded with the
+[LeRobot](https://github.com/huggingface/lerobot) policy classes.
+
+## Usage
+
+Clone the inference repo, then load this checkpoint directly from the Hub:
+
+```bash
+git clone https://github.com/yanweiw/itps.git && cd itps
+pip install -e .
+python interact_maze2d.py -p act --hf
+```
+
+Or load it programmatically:
+
+```python
+from itps.common.policies.act.modeling_act import ACTPolicy
+
+policy = ACTPolicy.from_pretrained("felixw/itps-act")
+policy.eval()
+```
+
+## Citation
+
+```bibtex
+@article{wang2024itps,
+  title={Inference-Time Policy Steering through Human Interactions},
+  author={Wang, Yanwei and others},
+  journal={arXiv preprint arXiv:2411.16627},
+  year={2024}
+}
+```
+
+## License
+
+MIT — see [LICENSE](https://github.com/yanweiw/itps/blob/main/LICENSE).

config.json

@@ -0,0 +1,41 @@
+{
+  "chunk_size": 64,
+  "dim_feedforward": 3200,
+  "dim_model": 512,
+  "dropout": 0.1,
+  "feedforward_activation": "relu",
+  "input_normalization_modes": {
+    "observation.environment_state": "mean_std",
+    "observation.state": "mean_std"
+  },
+  "input_shapes": {
+    "observation.environment_state": [
+      2
+    ],
+    "observation.state": [
+      2
+    ]
+  },
+  "kl_weight": 10.0,
+  "latent_dim": 32,
+  "n_action_steps": 64,
+  "n_decoder_layers": 1,
+  "n_encoder_layers": 4,
+  "n_heads": 8,
+  "n_obs_steps": 1,
+  "n_vae_encoder_layers": 4,
+  "output_normalization_modes": {
+    "action": "mean_std"
+  },
+  "output_shapes": {
+    "action": [
+      2
+    ]
+  },
+  "pre_norm": false,
+  "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
+  "replace_final_stride_with_dilation": false,
+  "temporal_ensemble_coeff": null,
+  "use_vae": true,
+  "vision_backbone": "resnet18"
+}

config.yaml

@@ -0,0 +1,173 @@
+resume: false
+device: cuda:0
+use_amp: false
+seed: 100000
+dataset_repo_id: maze2d
+dataset_root: /afs/csail.mit.edu/u/f/felixw/lerobot/data/maze2d-large-sparse-v1.hdf5
+video_backend: pyav
+training:
+  offline_steps: 100000
+  num_workers: 4
+  batch_size: 256
+  eval_freq: 0
+  log_freq: 250
+  save_checkpoint: true
+  save_freq: 5000
+  online_steps: 0
+  online_rollout_n_episodes: 1
+  online_rollout_batch_size: 1
+  online_steps_between_rollouts: 1
+  online_sampling_ratio: 0.5
+  online_env_seed: null
+  online_buffer_capacity: null
+  online_buffer_seed_size: 0
+  do_online_rollout_async: false
+  image_transforms:
+    enable: false
+    max_num_transforms: 3
+    random_order: false
+    brightness:
+      weight: 1
+      min_max:
+      - 0.8
+      - 1.2
+    contrast:
+      weight: 1
+      min_max:
+      - 0.8
+      - 1.2
+    saturation:
+      weight: 1
+      min_max:
+      - 0.5
+      - 1.5
+    hue:
+      weight: 1
+      min_max:
+      - -0.05
+      - 0.05
+    sharpness:
+      weight: 1
+      min_max:
+      - 0.8
+      - 1.2
+  lr: 1.0e-05
+  lr_backbone: 1.0e-05
+  weight_decay: 0.0001
+  grad_clip_norm: 10
+  delta_timestamps:
+    action:
+    - 0.0
+    - 0.1
+    - 0.2
+    - 0.3
+    - 0.4
+    - 0.5
+    - 0.6
+    - 0.7
+    - 0.8
+    - 0.9
+    - 1.0
+    - 1.1
+    - 1.2
+    - 1.3
+    - 1.4
+    - 1.5
+    - 1.6
+    - 1.7
+    - 1.8
+    - 1.9
+    - 2.0
+    - 2.1
+    - 2.2
+    - 2.3
+    - 2.4
+    - 2.5
+    - 2.6
+    - 2.7
+    - 2.8
+    - 2.9
+    - 3.0
+    - 3.1
+    - 3.2
+    - 3.3
+    - 3.4
+    - 3.5
+    - 3.6
+    - 3.7
+    - 3.8
+    - 3.9
+    - 4.0
+    - 4.1
+    - 4.2
+    - 4.3
+    - 4.4
+    - 4.5
+    - 4.6
+    - 4.7
+    - 4.8
+    - 4.9
+    - 5.0
+    - 5.1
+    - 5.2
+    - 5.3
+    - 5.4
+    - 5.5
+    - 5.6
+    - 5.7
+    - 5.8
+    - 5.9
+    - 6.0
+    - 6.1
+    - 6.2
+    - 6.3
+eval:
+  n_episodes: 50
+  batch_size: 50
+  use_async_envs: false
+wandb:
+  enable: true
+  disable_artifact: false
+  project: lerobot
+  notes: ''
+fps: 10
+env:
+  name: maze2d
+  task: null
+  state_dim: 2
+  action_dim: 2
+  fps: ${fps}
+policy:
+  name: act
+  n_obs_steps: 1
+  chunk_size: 64
+  n_action_steps: 64
+  input_shapes:
+    observation.environment_state:
+    - 2
+    observation.state:
+    - ${env.state_dim}
+  output_shapes:
+    action:
+    - ${env.action_dim}
+  input_normalization_modes:
+    observation.environment_state: mean_std
+    observation.state: mean_std
+  output_normalization_modes:
+    action: mean_std
+  vision_backbone: resnet18
+  pretrained_backbone_weights: ResNet18_Weights.IMAGENET1K_V1
+  replace_final_stride_with_dilation: false
+  pre_norm: false
+  dim_model: 512
+  n_heads: 8
+  dim_feedforward: 3200
+  feedforward_activation: relu
+  n_encoder_layers: 4
+  n_decoder_layers: 1
+  use_vae: true
+  latent_dim: 32
+  n_vae_encoder_layers: 4
+  temporal_ensemble_coeff: null
+  dropout: 0.1
+  kl_weight: 10.0

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1463f404e77cb9716fe1157180408925de84d13ad2e51552f8f1ae9b796d6c05
+size 160723088