Openvla 7b Oft Finetuned Libero Spatial

🚀 微調視覺-語言-動作模型：優化速度與成功率

本項目聚焦於視覺-語言-動作模型的微調，旨在優化模型運行速度並提高任務成功率。本倉庫包含了適用於LIBERO-Spatial的OpenVLA-OFT檢查點，相關內容詳見論文 Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success。OpenVLA-OFT通過採用優化的微調技術，顯著提升了基礎OpenVLA模型的性能。

項目頁面：https://openvla-oft.github.io/

代碼倉庫：https://github.com/openvla-oft/openvla-oft

其他OpenVLA-OFT檢查點：https://huggingface.co/moojink?search_models=oft

🚀 快速開始

此示例展示瞭如何使用預訓練的OpenVLA-OFT檢查點生成動作塊。請確保你已按照GitHub README中的說明設置好conda環境。

import pickle
from experiments.robot.libero.run_libero_eval import GenerateConfig
from experiments.robot.openvla_utils import get_action_head, get_processor, get_proprio_projector, get_vla, get_vla_action
from prismatic.vla.constants import NUM_ACTIONS_CHUNK, PROPRIO_DIM

# 實例化配置（有關定義，請參閱 experiments/robot/libero/run_libero_eval.py 中的 GenerateConfig 類）
cfg = GenerateConfig(
    pretrained_checkpoint = "moojink/openvla-7b-oft-finetuned-libero-spatial",
    use_l1_regression = True,
    use_diffusion = False,
    use_film = False,
    num_images_in_input = 2,
    use_proprio = True,
    load_in_8bit = False,
    load_in_4bit = False,
    center_crop = True,
    num_open_loop_steps = NUM_ACTIONS_CHUNK,
    unnorm_key = "libero_spatial_no_noops",
)

# 加載 OpenVLA-OFT 策略和輸入處理器
vla = get_vla(cfg)
processor = get_processor(cfg)

# 加載 MLP 動作頭以生成連續動作（通過 L1 迴歸）
action_head = get_action_head(cfg, llm_dim=vla.llm_dim)

# 加載本體感受投影儀以將本體感受映射到語言嵌入空間
proprio_projector = get_proprio_projector(cfg, llm_dim=vla.llm_dim, proprio_dim=PROPRIO_DIM)

# 加載示例觀察：
#   observation (dict): {
#     "full_image": 主第三人稱圖像,
#     "wrist_image": 腕載相機圖像,
#     "state": 機器人本體感受狀態,
#     "task_description": 任務描述,
#   }
with open("experiments/robot/libero/sample_libero_spatial_observation.pkl", "rb") as file:
    observation = pickle.load(file)

# 生成機器人動作塊（未來動作序列）
actions = get_vla_action(cfg, vla, processor, observation, observation["task_description"], action_head, proprio_projector)
print("Generated action chunk:")
for act in actions:
    print(act)

📄 許可證

本項目採用MIT許可證。

📚 引用

如果你在研究中使用了本項目，請引用以下論文：

@article{kim2025fine,
  title={Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success},
  author={Kim, Moo Jin and Finn, Chelsea and Liang, Percy},
  journal={arXiv preprint arXiv:2502.19645},
  year={2025}
}