only one environment spawns in the simulation, even though --num_envs 10

[SuddenCucumber]

A little bit of context: i am trying to set up RL for a quadruped robot. My direct environments consist of a ground plane, the robot, and light.

Problem: I set --num_envs 10, and when using print(env_ids), as well as printing xyz position, i get a reading on all 10 environments. However, in the 3D simulation, only one environment is loaded in, and one robot is for some reason "half-buried" in the ground plane at 0,0,0. Currently, env_spacing = 1.0, and changing it makes no difference.

Here is the image of that single robot and the "half-buried" robot, which is also missing its legs, for some reason. It is visible from the right of the image that there is only one environment running in the simulation, meaning that isaaclab creates the other environments and somehow does not load them into the simulation (i don't know how this works, so i am hypothesizing).

I noticed that when setting clone_in_fabric = False and (optionally) replicate_physics = False, this issue seems to go away. For my purposes, however, I want to have clone_in_fabric = True and replicate_physics = True as well, besides the cartpole tutorial has them set as such and does not have problems with cloning environments.

any help would be appreciated

[RandomOakForest]

With clone_in_fabric=True and replicate_physics=True, you must follow the same cloning pattern as the Cartpole direct tutorial (spawn once, then let InteractiveScene clone), otherwise you get the “half-buried at 0,0,0” artifact you’re seeing.¹²³

How cloning is supposed to work

When replicate_physics=True, InteractiveScene treats your scene as a homogeneous env and clones the entire env (xforms + physics) from env 0 to the remaining envs.⁴¹
With clone_in_fabric=True, this cloning is done via Fabric for speed, which means you must not manually spawn per-env robots or manually offset them; the cloner will apply env_spacing and env origins.⁵¹

For a direct RL env, the recommended pattern is exactly what Cartpole uses:

@configclass
class MyQuadrupedEnvCfg(DirectRLEnvCfg):
    # robot(s)
    robot_cfg: ArticulationCfg = MY_QUAD_CFG.replace(
        prim_path="/World/envs/env_.*/Robot"   # note regex
    )

    # scene
    scene: InteractiveSceneCfg = InteractiveSceneCfg(
        num_envs=10,
        env_spacing=4.0,
        replicate_physics=True,
        clone_in_fabric=True,
    )

And in the env:

class MyQuadrupedEnv(DirectRLEnv):
    cfg: MyQuadrupedEnvCfg

    def _setup_scene(self):
        # spawn robot ONCE in env_0 via Articulation, then let scene clone
        self.robot = Articulation(self.cfg.robot_cfg)

        # ground plane only once at /World/ground
        spawn_ground_plane(prim_path="/World/ground", cfg=GroundPlaneCfg())

        # let InteractiveScene perform the cloning; do NOT do your own per-env cloning
        self.scene.clone_environments(copy_from_source=False)

        # no manual env-spacing here; env origins come from scene.env_origins

This is exactly the structure in the official Cartpole direct tutorial, where they use num_envs=4096, env_spacing=4.0, replicate_physics=True, clone_in_fabric=True and rely on clone_environments/env_origins rather than calling set_world_poses manually.²³

Why your workaround “fixes” it

• With clone_in_fabric=False and/or replicate_physics=False, InteractiveScene falls back to per-asset cloning and per-object xform duplication, which is more forgiving if you are manually spawning multiple assets or moving root poses yourself.⁴
• When you enable both flags, the system assumes a clean, homogeneous env 0; any extra robot at /World/robot or manual per-env spawn can result in a stray articulation sitting at world origin, intersecting the plane, and not being cloned correctly.¹⁴

So the behavior you see (one “good” env plus one buried robot) usually indicates:

• You spawned the quadruped once at /World/quad or similar, not under /World/envs/env_0/....
• Or you are calling set_world_poses with absolute world positions while scene.env_origins already encode env offsets, effectively double-applying or misapplying offsets in the Fabric-cloned setup.⁴²

What to change in your quadruped env

To keep clone_in_fabric=True and replicate_physics=True and have multiple visible envs:

1. Prim path must be env-regex based

Ensure your ArticulationCfg uses an env regex like in Cartpole:

robot_cfg: ArticulationCfg = MY_QUAD_CFG.replace(
    prim_path="/World/envs/env_0/Robot"   # WRONG: only env_0
)
# instead:
robot_cfg: ArticulationCfg = MY_QUAD_CFG.replace(
    prim_path="/World/envs/env_.*/Robot"  # CORRECT
)

This tells the cloner that the robot lives under an env root, and it will generate /World/envs/env_0/Robot, /World/envs/env_1/Robot, ….⁶²
2. Spawn once, then clone via scene

In _setup_scene, do:

def _setup_scene(self):
    self.robot = Articulation(self.cfg.robot_cfg)
    spawn_ground_plane(prim_path="/World/ground", cfg=GroundPlaneCfg())
    self.scene.clone_environments(copy_from_source=False)

Do not manually spawn or clone robots per env when replicate_physics=True.²⁴
3. Use default root states + env_origins in reset

Follow the Cartpole pattern for initial pose:

def _reset_idx(self, env_ids: Sequence[int] | None):
    if env_ids is None:
        env_ids = self.robot._ALL_INDICES
    super()._reset_idx(env_ids)

    default_root_state = self.robot.data.default_root_state[env_ids]
    # IMPORTANT: shift by env origins; do NOT hard-code spacings
    default_root_state[:, :3] += self.scene.env_origins[env_ids]

    self.robot.write_root_pose_to_sim(default_root_state[:, :7], env_ids)
    self.robot.write_root_velocity_to_sim(default_root_state[:, 7:], env_ids)

This ensures each clone gets the proper per-env offset and avoids all robots collapsing at (0,0,0).²

Footnotes

1. https://isaac-sim.github.io/IsaacLab/main/source/api/lab/isaaclab.scene.html ↩ ↩² ↩³ ↩⁴

2. https://isaac-sim.github.io/IsaacLab/main/source/tutorials/03_envs/create_direct_rl_env.html ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

3. https://github.com/isaac-sim/IsaacLab/blob/main/source/isaaclab_tasks/isaaclab_tasks/direct/cartpole/cartpole_env.py ↩ ↩²

4. https://isaac-sim.github.io/IsaacLab/main/source/api/lab/isaaclab.scene.html ↩ ↩² ↩³ ↩⁴ ↩⁵

5. https://isaac-sim.github.io/IsaacLab/main/source/how-to/optimize_stage_creation.html ↩

6. https://isaac-sim.github.io/IsaacLab/main/source/setup/walkthrough/api_env_design.html ↩