Action Chunking with Transformer (ACT) の再実装
pip install git+https://github.com/keio-crl/ACT.git
import torch
from act import ACT
batch_size, action_size = 8, 14
chunk_size = 50
act = 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]