About get_local_poses and get_world_poses

[ZisongXu]

Codes Below

def post_reset(self) -> None:
      # Monitor all robots in all envs
      # Match robots in all envs ===> "/World/envs/env_.*/franks"
      self.robots_view = Articulation(prim_paths_expr="/World/envs/env_.*/franka")
      # Monitor all objects in all envs
      # Match objects in all envs ===> "/World/envs/env_.*/object_.*"
      self.objects_view = RigidPrim(prim_paths_expr="/World/envs/env_.*/objects_.*")
      # Initialize
      self._scene.add(self.robots_view)
      self._scene.add(self.objects_view)
      self.reset()

def reset(self) -> None:
      total_objects = self.objects_view.count
      # Robot World Pos[num_envs, 3]
      robot_world_pos, _ = self.robots_view.get_world_poses()
      robot_local_pos, _ = self.robots_view.get_local_poses()
      print(robot_world_pos)
      print(robot_local_pos)

I found that the results of robot_world_pos and robot_local_pos are the same. I don't know what happened here. Are there any bugs in the code? Please help me. Thanks a lot!

[vick-yu]

Hi @ZisongXu,

This is expected behavior based on your scene hierarchy setup. Let me explain what's happening.

Understanding the Difference

• get_world_poses(): Returns poses relative to the world's origin (/World)
• get_local_poses(): Returns poses relative to each prim's parent prim's frame

In your code, the robots are located at:

• /World/envs/env_0/franka
• /World/envs/env_1/franka
• etc.

The parent of each robot is the corresponding environment prim (/World/envs/env_0, /World/envs/env_1, etc.).

Why They're Identical

If your environment prims (env_0, env_1, etc.) have no transforms applied (i.e., they're at the world origin with identity transforms), then:

local_pose_relative_to_parent = world_pose

This happens because the parent's transform is the identity transform, so converting from world space to parent space doesn't change the values.

How to Verify

Add this diagnostic code to check your parent transforms:

from pxr import UsdGeom, Usd
from isaacsim.core.utils.stage import get_current_stage

stage = get_current_stage()
for i in range(self._num_envs):
    parent_prim = stage.GetPrimAtPath(f"/World/envs/env_{i}")
    xform = UsdGeom.Xformable(parent_prim)
    transform = xform.ComputeLocalToWorldTransform(Usd.TimeCode.Default())
    print(f"env_{i} transform:\n{transform}")

If all transforms are identity matrices (essentially all zeros except 1's on the diagonal), that confirms why local and world poses are the same.

[ZisongXu]

Hi Dear @vick-yu :

Thanks for your reply!

So, you mean that get_local_pose returns the robot's coordinates relative to env0, env1, env2, etc., but env0, env1, env2, etc., are actually all located at the world origin (0,0,0). In other words, env0, env1, etc., are all overlapping, which is why get_local_pose and get_world_pose are the same.

[vick-yu]

Exactly! You've got it. 👍

To clarify one small detail: each env_0, env_1, env_2, etc. exists as a separate prim in the scene graph with different paths, but they all have identity transforms (position at [0, 0, 0], no rotation). So while they're distinct prims, their coordinate frames all align with the world origin.

In a typical Isaac Sim multi-environment setup, you'd use GridCloner to apply spatial separation to these environment prims. This spreads them out in 3D space so you can see all robots simultaneously without them physically overlapping.

Example with GridCloner

from isaacsim.core.utils.cloner import GridCloner
from pxr import UsdGeom

# Create grid cloner with 2-meter spacing
cloner = GridCloner(spacing=2.0)
cloner.define_base_env("/World/envs")
UsdGeom.Xform.Define(stage, "/World/envs/env_0")

# Add your Franka robot to env_0
add_reference_to_stage(
    usd_path=assets_root_path + "/Isaac/Robots/FrankaRobotics/FrankaPanda/franka.usd",
    prim_path="/World/envs/env_0/franka"
)

# Clone to create multiple environments
num_envs = 4
prim_paths = cloner.generate_paths("/World/envs/env", num_envs)
env_positions = cloner.clone(
    source_prim_path="/World/envs/env_0",
    prim_paths=prim_paths
)
# This will position environments at [0,0,0], [2,0,0], [4,0,0], [6,0,0]

Result After Using GridCloner

With a robot at local position [0, 0, 0] within each environment:

• get_world_poses() → [[0, 0, 0], [2, 0, 0], [4, 0, 0], [6, 0, 0]] (spread out in world)
• get_local_poses() → [[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]] (relative to each env's frame)

Now you'd see the difference clearly!

Your Current Setup

In your current setup without environment transforms:

• get_world_poses() → [[x, y, z], [x, y, z], [x, y, z], [x, y, z]] (all at same position)
• get_local_poses() → [[x, y, z], [x, y, z], [x, y, z], [x, y, z]] (same because parent is at origin)

This confirms everything is working as designed - you just need to add the spatial layout for your environments!

[vick-yu]

Hello!

As there haven’t been any recent responses, we’re closing this issue or discussion to help keep our forum organized.

If you’re still encountering this issue or have further questions, please feel free to open a new issue or discussion with updated details. When doing so, referencing or linking to this original discussion would be helpful for context.

Thank you for contributing to the NVIDIA Isaac Sim community!

Best regards,
The NVIDIA Isaac Sim Team