[Bug]: Genesis Manipulation Example Fails with CUDA JIT Linking Error #2089
Description
Bug Description
I can not run the grasp_train.py example from the Genesis/examples/manipulation.
ERROR:
(base) *****@*****:~/Genesis/examples/manipulation$ /home/*****/anaconda3/bin/python /home/*****/Genesis/examples/manipulation/grasp_train.py
[Genesis] [21:26:44] [WARNING] (MJCF) Approximating tendon by joint actuator for `finger_joint1`
[Genesis] [21:26:44] [WARNING] (MJCF) Actuator control gain and bias parameters cannot be reduced to a unique PD control position gain. Using max between gain and bias for joint `finger_joint1`.
[Genesis] [21:26:44] [WARNING] (MJCF) Approximating tendon by joint actuator for `finger_joint2`
[Genesis] [21:26:44] [WARNING] (MJCF) Actuator control gain and bias parameters cannot be reduced to a unique PD control position gain. Using max between gain and bias for joint `finger_joint2`.
[Genesis] [21:26:45] [WARNING] Reference robot position exceeds joint limits.
[Genesis] [21:26:45] [WARNING] Constraint solver time constant should be greater than 2*substep_dt. timeconst is changed from `0.005` to `0.01`). Decrease simulation timestep or increase timeconst to avoid altering the original value.
Using rasterizer
All Vulkan layers supported
CUDA linking Failed!
ERROR 4 in nvvmAddNVVMContainerToProgram, may need newer version of nvJitLink library
Error at /project/src/mw/cuda_exec.cpp:734 in auto madrona::compileCode(const char **, int64_t, const char **, int64_t, const char **, int64_t, const char **, int64_t, const MegakernelConfig *, int64_t, CompileConfig::OptMode, ExecutorMode, bool)::(anonymous class)::operator()(nvJitLinkResult) const
nvJitLink error: Internal error
已中止 (核心已转储)
Steps to Reproduce
python /home/*****/Genesis/examples/manipulation/grasp_train.py
Expected Behavior
The manipulation example should start running normally.
Environment
OS: Ubuntu 24.04
Python: 3.13.5
PyTorch: 2.9.1+cu126
CUDA: 13.0
GPU: NVIDIA 4060
Genesis version: installed via pip install genesis-world