Segmentation fault when evaluating in simpler_env

[Agenda8]

Hello,

Thank you for your great work. I encounter a bug when evaluating CogACT in simpler_env.

When running evaluation with CogACT, the process consistently crashes with Segmentation fault (core dumped) after completing two episodes. I use NVIDIA GeForce RTX 3090 Ti to run the evaluation.

The log output is as below. I have already dismissed the output of every step. The evaluation run in maniskill2_evaluator.py

Control mode:  arm_pd_ee_delta_pose_align_interpolate_by_planner_gripper_pd_joint_target_delta_pos_interpolate_by_planner
[2025-10-14 16:05:55.419] [svulkan2] [error] GLFW error: X11: Failed to open display 
[2025-10-14 16:05:55.425] [svulkan2] [warning] Continue without GLFW.
/home/jianrd/miniconda3/envs/cogact_sim/lib/python3.10/site-packages/gymnasium/core.py:311: UserWarning: WARN: env.is_final_subtask to get variables from other wrappers is deprecated and will be removed in v1.0, to get this variable you can do `env.unwrapped.is_final_subtask` for environment variables or `env.get_wrapper_attr('is_final_subtask')` that will search the reminding wrappers.
  logger.warn(
/home/jianrd/miniconda3/envs/cogact_sim/lib/python3.10/site-packages/gymnasium/core.py:311: UserWarning: WARN: env.get_language_instruction to get variables from other wrappers is deprecated and will be removed in v1.0, to get this variable you can do `env.unwrapped.get_language_instruction` for environment variables or `env.get_wrapper_attr('get_language_instruction')` that will search the reminding wrappers.
  logger.warn(
open top drawer
/home/jianrd/miniconda3/envs/cogact_sim/lib/python3.10/site-packages/gymnasium/core.py:311: UserWarning: WARN: env.robot_uid to get variables from other wrappers is deprecated and will be removed in v1.0, to get this variable you can do `env.unwrapped.robot_uid` for environment variables or `env.get_wrapper_attr('robot_uid')` that will search the reminding wrappers.
  logger.warn(
place apple into top drawer
[2025-10-14 16:07:05.518] [svulkan2] [error] GLFW error: X11: Failed to open display 
[2025-10-14 16:07:05.518] [svulkan2] [warning] Continue without GLFW.
open top drawer
place apple into top drawer
simpler_env/policies/sim_cogact/scripts/cogact_put_in_drawer_visual_matching.sh: line 23: 4170478 Segmentation fault     (core dumped) CUDA_VISIBLE_DEVICES=${gpu_id} python simpler_env/main_inference.py --policy-model cogact --ckpt-path ${ckpt_path} --robot google_robot_static --control-freq 3 --sim-freq 513 --max-episode-steps 200 --env-name ${env_name} --scene-name dummy_drawer --robot-init-x 0.652 0.652 1 --robot-init-y 0.009 0.009 1 --robot-init-rot-quat-center 0 0 0 1 --robot-init-rot-rpy-range 0 0 1 0 0 1 0 0 1 --obj-init-x-range -0.08 -0.02 3 --obj-init-y-range -0.02 0.08 3 --rgb-overlay-path ./ManiSkill2_real2sim/data/real_inpainting/open_drawer_b0.png ${EXTRA_ARGS}

[xuanlinli17]

Hi, have you tried the debugging section in readme?

[xuanlinli17]

Also cuda13 is probably problematic if you are using it.

[Agenda8]

Hi, have you tried the debugging section in readme?

Yes, I have tried vulkaninfo and it shows the information about Vulkan correctly.

Also cuda13 is probably problematic if you are using it.

My cuda version is 12.1.

[fengdelin]

pip install numpy==1.24.4

[forsko]

Hi, I also encountered the same problem. Could you please tell me if you have solved it?

[Agenda8]

I've been busy with other work lately and haven't had time to address this issue yet. I have tried before

pip install numpy==1.24.4

but still didn't solve the problem.

[forsko]

I've been busy with other work lately and haven't had time to address this issue yet. I have tried before

pip install numpy==1.24.4

but still didn't solve the problem.
Thank you for your reply. I have also tried numpy, but it still doesn't work. It's really frustrating, but thank you very much all the same!

[forsko]

Hello! Could you please tell me which server you used during the assessment - the local server or the online server? I'm wondering if this kind of problem only occurs with online servers.

[forsko]

Hi! My issue occurred when the CUDA version of the graphics card driver was 13.0. I tried using a server with a graphics card driver version of 12.4, and it successfully ran.