ACT

Action Chunking with Transformer (ACT) の再実装

サンプルコードはこちら >>

インストール

pip install git+https://github.com/keio-crl/ACT.git

API

importtorchfromactimportACTbatch_size, action_size=8, 14chunk_size=50act=ACT(action_size=action_size, chunk_size=chunk_size, ...)
action_prediction=act.inference_step(
slave_proprio=torch.rand(batch_size, action_size),
observation=torch.rand(batch_size, 3, 224, 224),
) # [8, 50, 14]

参考文献

論文

• Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware
• DE⫶TR: End-to-End Object Detection with Transformers
• Visuo-Tactile Pretraing for Cable Plugging

コード

• thonyzaozh/act
• Shaka-Labs/act
• facebookresearch/detr
• Abraham190137/TactileACT