Environment Builder

Environment compilation is the step that turns the three independent components — scene, embodiment, and task — into a runnable Isaac Lab environment. ArenaEnvBuilder does this by collecting the partial configuration each component contributes and merging them into a single ManagerBasedRLEnvCfg.

ArenaEnvBuilder merges the Scene, Embodiment, and Task into a runnable ManagerBasedRLEnv.

environment = IsaacLabArenaEnvironment(
    name="manipulation_task",
    embodiment=embodiment,
    scene=scene,
    task=task,
)

env_builder = ArenaEnvBuilder(environment, ArenaEnvBuilderCfg())
env = env_builder.make_registered()

How it works

Each component (Scene, Embodiment, Task) exposes a set of get_*_cfg() methods that return its contribution to each Isaac Lab manager. The typical contributions of each component to each manager are tabulated below:

Isaac Lab Manager	Isaac Lab - Arena Component
	Scene	Embodiment	Task
Scene	assets, lights	robot, sensors	task-specific assets
Observations		proprioception, cameras	goal observations
Actions		control interface
Events (resets)	object placement	robot reset	task reset
Terminations			success, failure
Rewards			dense rewards (RL)
Recorder			metrics-required data

ArenaEnvBuilder.compose_manager_cfg() first assembles the partial manager contributions from each component into a set of complete managers. Then it merges these complete managers into a single ManagerBasedRLEnvCfg.

By default, the builder also solves spatial relations for placed objects and supported robot embodiments. Set ArenaEnvBuilderCfg(solve_relations=False) in Python to disable this step. Arena command-line runners expose the equivalent --no_solve_relations option.

Seeds during compilation

Environment compilation and rollout use more than one random stream, so locking a single global seed is not enough when you need layouts, object-set picks, or run-time variation draws to be reproducible independently.

What it is

ArenaEnvBuilder exposes two seeds. They are independent — locking one does not fix the other.

Control	CLI / config	Default	Locking it reproduces
Environment seed	--seed / ArenaEnvBuilderCfg.seed	42	Simulation RNG after the Isaac Lab env is created: reset noise and run-time variation draws.
Placement seed	--placement_seed / ArenaEnvBuilderCfg.placement_seed	None (unlocked)	Relation-solver layouts and random RigidObjectSet member assignment. With None, placement stays non-reproducible across runs.

There is no variation seed. Run-time variations follow --seed; build-time variations are drawn once at compile time and are not locked by either seed. See Variations and Pooled Placement.

How to set it

Pass the seed you want to lock on the runner CLI (or set the matching field on ArenaEnvBuilderCfg / placer_params):

python isaaclab_arena/evaluation/policy_runner.py \
  --policy_type zero_action \
  --seed 42 \
  --placement_seed 7 \
  --num_steps 100 \
  pick_and_place_maple_table

• Set --seed to fix simulation and run-time variation draws.

• Set --placement_seed to fix layouts and random object-set picks.

• Omit --placement_seed when placement should vary across runs.

The compiled config is then registered with the gym registry under the environment’s name, and gym.make() returns the gym environment.

Mimic mode

Passing --mimic at the command line compiles a ManagerBasedRLMimicEnv instead of a standard ManagerBasedRLEnv. The mimic environment is used for demonstration generation and includes subtask configurations from the task. Metrics and recorders are excluded in mimic mode.

python submodules/IsaacLab/scripts/imitation_learning/isaaclab_mimic/generate_dataset.py \
  --external_callback isaaclab_arena.environments.isaaclab_interop.environment_registration_callback \
  --mimic ...

Next Steps

Continue to Relations and Strategies to learn how anchors and spatial relations describe a placement layout.