Environment Configuration
Configure environments with Hydra.
Command-Line Overrides
srb agent <WORKFLOW> --env <ENV> \
env.domain=moon \
env.robot=ur10 \
env.num_envs=4 \
env.stack=false \
...
Configuration Files
Each environment’s @configclass extends BaseEnvCfg. Hydra loads task Python defaults, merges CLI overrides, and writes the resolved config to <logdir>/.hydra/config.yaml unless the command disables its output directory.
Save overrides in YAML and pass them with --cfg:
srb agent zero --env peg_in_hole --cfg ./my_overrides.yaml
For example, my_overrides.yaml can contain:
env:
domain: mars
num_envs: 8
robot: ur10
Note:
--cfg defaultuses the environment default;--cfg ignore,none, ornulldisables the config file.
Tip: Dict-form asset shortcuts expand automatically:
env.robot=frankabecomesenv.robot.name=franka.
Key Parameters
Scenario/Environment
env.domain- Domainenv.ephemeris- Ephemerisenv.robot- Robotenv.scenery- Set a registry asset name, such asenv.scenery=moon_surface.- For asset-instance scenery, set fields such as
env.scenery.flat_area_size=3.0orenv.scenery.boulder_count=0. - If the task accepts a scenery variant (
Scenery | AssetVariant), you can also setbody,lat_deg,lon_deg, andsize_m. Setlat_deg,lon_deg, andsize_mto constructRealTerrain. The defaultbodyisenv.domain(moon,mars, orearth). - Fixed-type scenery, such as
terrain_landscaping’sPlane, accepts only that asset’s fields.
- For asset-instance scenery, set fields such as
env.robot_spawn_height_range- Ground tasks only: root height offset range in metres above the robot’s default spawn height, applied on reset, e.g.env.robot_spawn_height_range=[0.3,0.5]. If unset, the task’srandomize_robot_staterange applies.env.num_envs,env.stack- Parallelismenv.particles- Particlesenv.reward- Reward & Normalization
Simulation
env.sim- Low-level simulation parameters (physics, rendering, etc.)env.visuals- Visual appearance settings
Debugging
env.debug_vis- Enables debug visualization features