Environment Configuration — Parallelism

Set env.num_envs to choose the number of parallel simulation instances.

Number of Environments

env.scene.num_envs sets the count; env.num_envs overrides it when set:

srb agent rand -e _manipulation env.scene.num_envs=16
srb agent zero -e _manipulation env.num_envs=128

Instances share one batched DirectEnv; cross-environment collisions are filtered. Visual sensors are not guaranteed isolation: spacing or sensor range alone may not prevent cross-environment visibility.

Environment Spacing

env.scene.env_spacing sets environment distance. env.spacing overrides it when set; otherwise the scene value is inherited:

srb agent rand -e _manipulation env.num_envs=64 env.scene.env_spacing=20.0
srb agent rand -e _manipulation env.num_envs=32 env.spacing=10.0

Note: env.stack=true forces env.scene.env_spacing=0.0; otherwise the framework applies env.spacing when set or inherits env.scene.env_spacing.

Environment Stacking

env.stack controls asset sharing:

Independent Environments (env.stack=false)

• ⚠️ Each environment has its own scene and position
• ✅ Greater visual and physical diversity
• ✅ Supports tasks with visual sensors
• ❗ Slower initialization
• ❗ Higher memory use

srb agent rand -e _manipulation env.num_envs=16 env.stack=false

Stacked Environments (env.stack=true)

• ⚠️ Environments share scenery and position
• ✅ Faster initialization
• ✅ Lower memory use
• ❗ Less environmental diversity
• ❗ Visual sensor compatibility and camera isolation are not guaranteed; verify the task before relying on them
• ❗ Stacked multi-environment particle simulation is unsupported: with env.particles=true and env.num_envs != 1, the framework forces env.stack=false and logs a warning

srb agent rand -e _manipulation env.num_envs=16 env.stack=true

Note: Set env.stack=false for visual observations or independent scenery. Defaults are task-specific. These task IDs use env.stack=true: excavation_demo, lunalab, velocity_tracking, locomotion_velocity_tracking, waypoint_navigation, locomotion_waypoint_navigation, orbital_waypoint_navigation, formation_following, formation_following_decentralized, formation_following_orbital, moonfall, traversal, and landing. Stacking shares scenery and forces spacing to zero:

srb agent train --algo dreamer --env waypoint_navigation \
  env.num_envs=64 env.stack=false