Custom Model Configs¶
The model configuration handles the following items:
1. Defines the joint action and observation limits
2. Defines joints as periodic or linear observations.
3. Defines task_mode (Which model is being used i.e fixed_hip, free_hip, etc)
4. Allows observation masking (Omit specified observation from observation space)
5. Defines robot dimensions. Used to calculate leg angles to match specific reset height
Warning
I do not recommend changing the model config file without a complete understanding of the projects software stack. Configuration files might be fragile and will be updated in future along with gym-os2r.
If you still want to update the model config here is a short explanation on
what to expect. Firstly It is recommended to be have gym-os2r installed as a
developer. The config implementation can be found here.
The default configuration file is found in gym_os2r/models/config/default/settings.yaml.
The example showing how to update or add reset positions here, shows how the configuration can be
updated externally (be careful, there aren’t many safe guards for bad modifications).
Alternatively you can update the settings.yaml file directly. This is the recommended method for larger
changes. The settings file has a little bit of documentation describing what changing each
variable represents.
Note
If you have a better idea on how to structure the model configuration
(joint information and reset information) or more generally gym-os2r I
would gladly attempt to implement/reformat the package structure.
Custom Reset Position¶
The information defining each reset position is below. Each item is defined as a xpath. Lay implies the robot will be horizontal in a ‘laying’ position. ‘planarizer_pitch_joint’ defines the angle of the planarizer pitch from horizontal.
resets:
name_reset:
laying_down: True
planarizer_pitch_joint: 0.4
We can set/modify the value for a new/existing reset position following the method below.
from gym_os2r import randomizers
from gym_os2r.common import make_env_from_id
from gym_os2r.models.config import SettingsConfig
reset_position = 'name_reset'
cfg = SettingsConfig()
xpath = 'resets/' + reset_position
new_pitch = 0.4
laying_down = True
cfg.set_config(laying_down, xpath + '/laying_down')
cfg.set_config(new_pitch, xpath + '/planarizer_pitch_joint')
env_id = "Monopod-stand-v1"
kwargs = {
'reset_positions': ['name_reset'],
'config': cfg
}
make_env = functools.partial(make_env_from_id, env_id=env_id, **kwargs)
env = randomizers.monopod_no_rand.MonopodEnvNoRandomizer(env=make_env)