Skip to content

Latest commit

 

History

History
60 lines (48 loc) · 2.56 KB

File metadata and controls

60 lines (48 loc) · 2.56 KB

Set up wandb

We use wandb to log out results. By default it is set to offline under the configuration file the tabletop or drawer. Before training anything please log into your wandb and change the values of wandb_login in the previous file.

Training

The models and hydra configuration will be stored under VAPO_ROOT/task/hydra_outputs/date/time If you want to log the outputs to wandb, you can add your credentials to the hydra configuration of either the tabletop or drawer experiments. The default configuration is set to run with the parameters of VAPO.

Tabletop

VAPO

python ./scripts/train_tabletop.py

Baseline

python ./scripts/train_tabletop.py affordance.gripper_cam.use_distance=False affordance.gripper_cam.use=False affordance.gripper_cam.densify_reward=False 

Generalization

sending the scene parameter with the name of any of the yaml files in ./config/scene will start the training with said configuration. For the generalization experiments we used the following:

VAPO

The default configuration is set to run with the parameters of VAPO

python ./scripts/train_tabletop.py scene=tabletop_random_unseen_15objs

Baseline

python ./scripts/train_tabletop.py model_name=baseline scene=tabletop_random_unseen_15objs affordance.gripper_cam.use_distance=False affordance.gripper_cam.use=False affordance.gripper_cam.densify_reward=False 

Drawer

VAPO

The default configuration is set to run with the parameters of VAPO

python ./scripts/train_playtable.py

Baseline

python ./scripts/train_playtable.py model_name=baseline affordance.gripper_cam.use_distance=False affordance.gripper_cam.use=False affordance.gripper_cam.densify_reward=False 

Testing

Adding the argument viz_obs=True will result in windows showing what the agent is seeing. To select the available scenes please see the list of the scene hydra configuration files. The camera configuration tabletop_render is the one used to produce the videos, which additionally to the robot observations includes one high resolution full-view render of the scene.

Tabletop

From the root directory run:

Vapo

python ./scripts/eval_tabletop.py viz_obs=true camera_conf=tabletop_render test.folder_name=./trained_models/policy/tabletop/vapo

Baseline

python ./scripts/eval_tabletop.py viz_obs=true camera_conf=tabletop_render test.folder_name=./trained_models/policy/tabletop/baseline