Skip to content

Crash in aloha pot env with SAP_TILE broadphase #458

@Kenny-Vilella

Description

@Kenny-Vilella

The aloha pot environment exhibits crash (segmentation fault).

For instance, this crash on my computer:
python mujoco_warp/testspeed.py --function=step --is_sparse=True --mjcf=benchmark/aloha_pot/scene.xml --nstep=10000 --ls_parallel=True --batch_size=1024 --nconmax=3000000 --njmax=400000

No crash happen when the broadphase is changed to NXN or SAP_SEGMENTED.
My preliminary investigation has shown that the crash occurs in _sap_range when accessing sap_projection_lower_in.
Compute sanitizer indicates an illegal read.
It looks like the array get corrupted when calling tile_sort in _segmented_sort.

The seg fault is still present without graph capture.
I noticed that it stopped to crash when I added some lines of code to save the array and print some results, but it still crashing when I added synchronization or and sleeping.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions