Hierarchical-ACT

J. Hyeon Park, Wonhyuk Choi, Sunpyo Hong, Hoseong Seo, Joonmo Ahn, Changsu Ha, Heungwoo Han, Junghyun Kwon, and Sungchul Kang "Hierarchical Action Chunk Transformer: Learning Temporal Multimodality from Demonstrations with Fast Imitation Behavior", IEEE International Conference on Intelligent Robots and Systems (IROS), 2024

Experiment video

Stack cups

Close pocket

Open lid

List of maintainers

• Jaehyeon Park ([email protected])
• Wonhyuk Choi ([email protected])

Installation

conda create -n robot_action_learner python=3.8
conda activate robot_action_learner
pip install -r requirements.txt

Train with sample data

python ./train.py \
    --exp_dir ./experiment \
    --data_root ./data_samples \
    --data_config ./configs/dataset/srrc_dual_frankas/stack_cups.gin \
    --model_config ./configs/model/hact_vq/srrc_dual_frankas/base.gin \
    --task_config ./configs/task/hact_vq/srrc_dual_frankas/base_local.gin \
    --gpu 0

• model.pt will be created in exp_dir.

Create agent

python ./serve.py --exp_dir ./experiment

• agent.pt will be created in exp_dir.

Deploy agent

agent = torch.jit.load("./experiment/agent.pt")
agent = agent.eval()

obs = env.reset()
while not done:
    action = agent.forward(obs, timestep=t)
    obs, reward, done, truncated, info = env.step(action)

Todo

• We will open the simulation (Viper-X) and datasets soon.

Code information

This code is implemented on the top of ACT. Several codes are adopted or modified from ACT.

Acknowledgement

All members of our remarkable robotics team: Joonmo Ahn, Rakjoon Chung, Changsu Ha, Heungwoo Han, Sunpyo Hong, Jaesik Jang, Rijeong Kang, Hosang Lee, Dongwoo Park, Hoseong Seo, Jaemin Yoon (in alphabetic order)