Fixes physx replication and scene query performance issues by kellyguo11 · Pull Request #4791 · isaac-sim/IsaacLab

kellyguo11 · 2026-03-02T19:36:42Z

Description

Fixes issue with PhysX cloning replication logic where it was causing overflows of GPU buffers in PhysX as collisions across environments were not being filtered correctly. Adds back the scene query flag that was not being correctly set into the physics scene, which also caused a performance regression.

Fixes UDIM parsing in the asset download logic.

Type of change

Bug fix (non-breaking change which fixes an issue)

Checklist

I have read and understood the contribution guidelines
I have run the pre-commit checks with ./isaaclab.sh --format
I have made corresponding changes to the documentation
My changes generate no new warnings
I have added tests that prove my fix is effective or that my feature works
I have updated the changelog and the corresponding version in the extension's config/extension.toml file
I have added my name to the CONTRIBUTORS.md or my name already exists there

…-issues

greptile-apps · 2026-03-02T19:47:55Z

Greptile Summary

This PR fixes two PhysX simulation correctness/performance regressions and a UDIM asset-download bug.

PhysX replication fix (physx_replicate.py): attach_fn now excludes both /World/template and /World/envs so PhysX does not independently parse environment prims; the replicator owns all bodies under /World/envs. Self-replication is excluded by default (exclude_self_replication=True) when a source maps to other environments, preventing GPU buffer overflows caused by duplicate collision pairs. A num_envs > 1 guard skips registration entirely for single-environment setups where normal USD parsing is sufficient. As an acknowledged trade-off, useEnvIds is hardcoded to False (with a TODO) until heterogeneous-source support is added.
Scene query support fix (physx_manager.py / physx_manager_cfg.py): Restores the enable_scene_query_support flag that was not being forwarded from SimulationCfg to the PhysX scene, recovering the associated raycasting/sweep performance that regressed when the flag defaulted to False.
UDIM download fix (assets.py): UDIM texture placeholders are now expanded to individual tile URLs (1001–1100) before downloading, preventing broken texture downloads.
Cleanup: Several task configs that had replicate_physics = False as a workaround have that override removed now that replication works correctly. The Dexsuite env no longer needs oversized GPU aggregate-pair buffers that compensated for the collision-filtering overflow.

Issues found:

prettytable==3.3.0 was removed from isaaclab_newton/setup.py but PrettyTable is still imported in isaaclab_newton/assets/articulation/articulation.py, which will cause a runtime ImportError on fresh installs.
The UDIM tile expansion stops at the first missing tile (break), silently skipping any tiles after a gap in a non-contiguous UDIM set.
useEnvIds=False is a known performance regression for GPU multi-env simulations (TODO comment present).

Confidence Score: 3/5

Safe to merge for simulation correctness, but the prettytable removal will cause a runtime ImportError for isaaclab_newton users and should be resolved first.
The core PhysX replication and scene-query fixes are well-reasoned and backed by updated tests. However, dropping prettytable from INSTALL_REQUIRES while it is still imported in articulation.py is a clear packaging regression that will break isaaclab_newton on fresh installs. The useEnvIds=False trade-off is acknowledged technical debt. These lower the score from an otherwise solid bug-fix PR.
source/isaaclab_newton/setup.py — prettytable removed but still used in articulation.py. source/isaaclab_physx/isaaclab_physx/cloner/physx_replicate.py — useEnvIds=False performance TODO needs follow-up tracking.

Important Files Changed

Filename	Overview
source/isaaclab_physx/isaaclab_physx/cloner/physx_replicate.py	Core fix: adds `/World/envs` to `attach_fn` exclusion list, pre-computes effective world lists with self-exclusion, and wraps all registration in a `num_envs > 1` guard. `useEnvIds` is hardcoded to `False` with a TODO, which is a known performance trade-off.
source/isaaclab/isaaclab/utils/assets.py	Adds UDIM placeholder expansion via tile-probing (1001–1100); assumes contiguous tile numbering and stops at the first missing tile. Non-contiguous UDIM sets would be partially downloaded.
source/isaaclab_physx/isaaclab_physx/physics/physx_manager.py	Restores `enable_scene_query_support` forwarding from `SimulationCfg` to the PhysX scene and overrides to `True` when the GUI is active. The value is correctly applied via the generic `cfg.to_dict()` loop.
source/isaaclab_physx/isaaclab_physx/physics/physx_manager_cfg.py	Adds `enable_scene_query_support: bool = False` field with clear docstring; default is `False` for headless performance with GUI override logic documented.
source/isaaclab_newton/setup.py	Removes `prettytable==3.3.0` from INSTALL_REQUIRES, but `prettytable` is still imported in `isaaclab_newton/assets/articulation/articulation.py`, which will cause a runtime ImportError.
source/isaaclab_physx/test/sim/test_cloner.py	Tests updated to reflect new exclude-self default, `num_envs > 1` guard, and `/World/envs` exclusion from `attach_fn`. The `test_physx_replicate_vs_no_replicate` xfail reason is updated to match the new replicator behavior.
source/isaaclab_tasks/isaaclab_tasks/manager_based/manipulation/dexsuite/dexsuite_env_cfg.py	Removes `gpu_found_lost_aggregate_pairs_capacity=229` and `gpu_total_aggregate_pairs_capacity=225` overrides that were workarounds for the collision-filtering overflow; now falls back to sensible defaults.
source/isaaclab_tasks/isaaclab_tasks/direct/automate/assembly_env.py	Captures the return value of `retrieve_file_path` instead of reconstructing the path with `os.path.basename`; cleaner and avoids path construction bugs.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant physx_replicate
    participant PhysXReplicatorInterface
    participant PhysXEngine

    Caller->>physx_replicate: physx_replicate(stage, sources, destinations, mapping, ...)

    alt num_envs == 1
        Note over physx_replicate: Skip replicator registration entirely.<br/>PhysX parses /World/envs normally.
    else num_envs > 1
        physx_replicate->>physx_replicate: Pre-compute effective_worlds[i]<br/>(exclude self when other envs exist)
        physx_replicate->>PhysXReplicatorInterface: register_replicator(stage_id, attach_fn, attach_end_fn, rename_fn)

        PhysXReplicatorInterface->>physx_replicate: attach_fn(stage_id)
        physx_replicate-->>PhysXReplicatorInterface: ["/World/template", "/World/envs"]
        Note over PhysXEngine: PhysX skips independent parsing<br/>of /World/template and /World/envs

        PhysXReplicatorInterface->>physx_replicate: attach_end_fn(stage_id)
        loop for each source i
            physx_replicate->>PhysXReplicatorInterface: rep.replicate(stage_id, src, len(worlds), useEnvIds=False)
            Note over PhysXReplicatorInterface: rename_fn called per copy:<br/>current_template.format(current_worlds[j])
        end
        physx_replicate->>PhysXReplicatorInterface: unregister_replicator(stage_id)
    end

_{Last reviewed commit: d6039b4}

greptile-apps

_{16 files reviewed, 4 comments}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-03-02T19:47:59Z

source/isaaclab_physx/isaaclab_physx/cloner/physx_replicate.py

    def attach_fn(_stage_id: int):
-        return ["/World/template"]
+        return ["/World/envs", *sources]


reverts the fix from PR #4731 which changed this to return ["/World/template"] to prevent bugs from cloning the source environment - verify this doesn't re-introduce those issues

ooctipus · 2026-03-02T19:51:47Z

source/isaaclab_physx/isaaclab_physx/physics/physx_manager.py

        )
        sim_utils.safe_set_attribute_on_usd_prim(scene_prim, "physxScene:enableGPUDynamics", is_gpu, camel_case=False)

+        # scene query support (from SimulationCfg, not PhysxCfg)


hmmm this may needs from physx

source/isaaclab/isaaclab/sim/simulation_context.py

Signed-off-by: Kelly Guo <kellyg@nvidia.com>

source/isaaclab_physx/isaaclab_physx/physics/physx_manager_cfg.py

Signed-off-by: Kelly Guo <kellyg@nvidia.com>

…ic into fix-more-issues

greptile-apps · 2026-03-02T23:02:26Z

source/isaaclab_newton/setup.py


 # Minimum dependencies required prior to installation
 INSTALL_REQUIRES = [
-    # generic
-    "prettytable==3.3.0",
    # newton
+    "warp-lang==1.12.0rc2",  # avoids pulling newton dep from newton package that could end up being a dev build
    "mujoco==3.5.0",
    "mujoco-warp==3.5.0",
-    "warp-lang==1.12.0rc2",
    "newton @ git+https://github.com/newton-physics/newton.git@v0.2.3",


prettytable removed but still used in articulation.py

prettytable==3.3.0 has been dropped from INSTALL_REQUIRES, but source/isaaclab_newton/isaaclab_newton/assets/articulation/articulation.py still contains:

from prettytable import PrettyTable

This will raise an ImportError at runtime for any user who installs isaaclab_newton without prettytable already present in their environment (e.g. a fresh install). Either add prettytable back to INSTALL_REQUIRES (or update it to allow a version range), or remove the import and its usages from articulation.py.

greptile-apps · 2026-03-02T23:02:27Z

source/isaaclab/isaaclab/utils/assets.py

+            if _UDIM_RE.search(cur_url):
+                for tile in range(1001, 1101):
+                    tile_url = _UDIM_RE.sub(str(tile), cur_url)
+                    if omni.client.stat(tile_url.replace(os.sep, "/"))[0] == omni.client.Result.OK:
+                        if tile_url not in visited:
+                            to_visit.append(tile_url)
+                    else:
+                        break
+                continue


UDIM tile expansion assumes perfectly contiguous tiles

The break on the first missing tile is correct for the common case, but UDIM sets can occasionally have gaps (e.g. tiles 1001–1002 exist, 1003 is absent, 1004–1006 exist). In that scenario tiles after the first gap would silently be skipped, leading to a partial download that only fails at render time.

Consider whether a non-contiguous walk (continue instead of break, stopping only at the cap) or at least a warning log would be safer:

# Option: warn instead of silently stopping else: logger.debug("UDIM tile %d missing for %s; stopping expansion.", tile, cur_url) break

If the project's assets are guaranteed to be contiguous this is fine to leave as-is, but documenting that assumption in the comment would help future maintainers.

greptile-apps · 2026-03-02T23:02:28Z

source/isaaclab_physx/isaaclab_physx/cloner/physx_replicate.py

+                    # TODO: envIds needs to support heterogeneous setup. for now, we rely on USD collision filtering
+                    useEnvIds=False,  # (len(current_worlds) == num_envs - 1) and device != "cpu",


useEnvIds=False disables GPU inter-env collision filtering optimisation

The previous logic used useEnvIds=True when the source covered all environments on CUDA, which lets PhysX use its fast GPU-side per-environment ID tracking to filter cross-env collisions. Hard-coding False falls back to USD collision filtering for every call, which should be correct but may introduce a performance regression on large GPU simulations.

The TODO comment captures this, so this is acknowledged debt — just flagging it to make sure it is tracked in the issue/PR backlog and not forgotten after the merge.

kellyguo11 added 5 commits March 2, 2026 10:09

fix scene query

ba7f1c7

Test more fixes

981ad71

fix udim parsing in assets

b1ac7c9

Merge branch 'develop' of github.com:isaac-sim/IsaacLab into fix-more…

0e9a9d8

…-issues

revert change in physx manager

9d73077

kellyguo11 requested review from Mayankm96, matthewtrepte and ossamaAhmed as code owners March 2, 2026 19:36

kellyguo11 marked this pull request as draft March 2, 2026 19:36

github-actions bot added bug Something isn't working isaac-lab Related to Isaac Lab team labels Mar 2, 2026

greptile-apps bot reviewed Mar 2, 2026

View reviewed changes

ooctipus reviewed Mar 2, 2026

View reviewed changes

kellyguo11 added 2 commits March 2, 2026 12:41

update physx replication

8082dac

revert physics replicate changes in mimic envs

2435bb5

kellyguo11 commented Mar 2, 2026

View reviewed changes

source/isaaclab/isaaclab/sim/simulation_context.py Outdated Show resolved Hide resolved

Apply suggestion from @kellyguo11

2ff2156

Signed-off-by: Kelly Guo <kellyg@nvidia.com>

kellyguo11 commented Mar 2, 2026

View reviewed changes

source/isaaclab_physx/isaaclab_physx/physics/physx_manager_cfg.py Outdated Show resolved Hide resolved

kellyguo11 and others added 6 commits March 2, 2026 12:47

Apply suggestion from @kellyguo11

7cfd8f2

Signed-off-by: Kelly Guo <kellyg@nvidia.com>

update scene query logic

25d37de

Merge branch 'fix-more-issues' of github.com:kellyguo11/IsaacLab-publ…

69a74af

…ic into fix-more-issues

pin warp dep

e4b1932

trigger CI

fb8beb7

fix CI

fe401d5

kellyguo11 changed the title ~~Fix more issues~~ Fixes physx replication and scene query performance issues Mar 2, 2026

kellyguo11 added 2 commits March 2, 2026 14:29

disable envIDs for now

9c38a4c

update license exceptions

d6039b4

kellyguo11 marked this pull request as ready for review March 2, 2026 22:47

kellyguo11 requested a review from hhansen-bdai as a code owner March 2, 2026 22:47

github-actions bot added the infrastructure label Mar 2, 2026

ooctipus approved these changes Mar 2, 2026

View reviewed changes

greptile-apps bot reviewed Mar 2, 2026

View reviewed changes

kellyguo11 merged commit 96e2948 into isaac-sim:develop Mar 2, 2026
6 of 11 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fixes physx replication and scene query performance issues#4791

Fixes physx replication and scene query performance issues#4791
kellyguo11 merged 16 commits intoisaac-sim:developfrom
kellyguo11:fix-more-issues

kellyguo11 commented Mar 2, 2026 •

edited

Loading

Uh oh!

greptile-apps bot commented Mar 2, 2026 •

edited

Loading

Uh oh!

greptile-apps bot left a comment •

edited

Loading

Uh oh!

greptile-apps bot Mar 2, 2026

Uh oh!

ooctipus Mar 2, 2026

Uh oh!

Uh oh!

Uh oh!

greptile-apps bot Mar 2, 2026

Uh oh!

greptile-apps bot Mar 2, 2026

Uh oh!

greptile-apps bot Mar 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		# TODO: envIds needs to support heterogeneous setup. for now, we rely on USD collision filtering
		useEnvIds=False, # (len(current_worlds) == num_envs - 1) and device != "cpu",

Conversation

kellyguo11 commented Mar 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Type of change

Checklist

Uh oh!

greptile-apps bot commented Mar 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 3/5

Important Files Changed

Sequence Diagram

Uh oh!

greptile-apps bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

ooctipus Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

greptile-apps bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Mar 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

kellyguo11 commented Mar 2, 2026 •

edited

Loading

greptile-apps bot commented Mar 2, 2026 •

edited

Loading

greptile-apps bot left a comment •

edited

Loading