Skip to content

Latest commit

 

History

History
55 lines (49 loc) · 2.08 KB

quick_start.md

File metadata and controls

55 lines (49 loc) · 2.08 KB
Raw

Quick start

This quick start allows you to collect data in the MuJoCo simulation and train and rollout the ACT policy.

Install

Install RoboManipBaselines:

$ git clone [email protected]:isri-aist/RoboManipBaselines.git --recursive
$ cd RoboManipBaselines
$ pip install -e .[act]

Install ACT from a third party:

$ cd third_party/act/detr
$ pip install -e .

Data collection by teleoperation

Note: Instead of collecting data by teleoperation, you can download the public dataset TeleopMujocoUR5eCable_Dataset30 from here.

Operate the robot in the simulation and save the data:

$ cd robo_manip_baselines/teleop
$ # Connect a SpaceMouse 3D mouse to your PC
$ python bin/TeleopMujocoUR5eCable.py --world_idx_list 0 5

In our experience, models can be trained stably with roughly 30 data sets. The teleoperation data is saved in the robo_manip_baselines/teleop/teleop_data/MujocoUR5eCable_<date_suffix> directory (e.g., MujocoUR5eCable_20240101_120000) in npz format.

Model training

Train the ACT:

$ cd robo_manip_baselines/act
$ python ../utils/make_dataset.py \
--in_dir ../teleop/teleop_data/MujocoUR5eCable_20240101_120000 \
--out_dir ./data/MujocoUR5eCable_20240101_120000 \
--train_ratio 0.8 --nproc `nproc` --skip 3
$ python ./bin/TrainAct.py \
--dataset_dir ./data/MujocoUR5eCable_20240101_120000 \
--log_dir ./log/MujocoUR5eCable_20240101_120000

Note: The following error will occur if the chunk_size is larger than the time series length of the training data. In such a case, either set the --skip option in make_dataset.py to a small value, or set the --chunk_size option in TrainAct.py to a small value.

RuntimeError: The size of tensor a (70) must match the size of tensor b (102) at non-singleton dimension 0

Policy rollout

Rollout the ACT in the simulation:

$ cd robo_manip_baselines/act
$ python ./bin/rollout/RolloutActMujocoUR5eCable.py \
--checkpoint ./log/MujocoUR5eCable_20240101_120000/policy_last.ckpt \
--skip 3 --world_idx 0