启动simtoreal后突然无法在无动作输入时，按a启动后无法站立，持续抽搐 #27
Description
以下时启动bash sim2real.sh的日志输出：
Starting TWIST2 real robot controller...
Policy file: /home/jqjs/workspace/TWIST2/assets/ckpts/twist2_1017_25k.onnx
Config file: robot_control/configs/g1.yaml
Device: cuda
Network interface: enp129s0
Use hand: True
Record proprio: False
Smooth body: 0.5
SAFETY WARNING:
You are about to run a policy on a real robot.
Make sure the robot is in a safe environment.
Press Ctrl+C to stop at any time.
Use the remote controller [Select] button to exit.
G1-HG type: 5
UnitreeInterface initialized: G1-HG (29 motors, HG messages) on interface: enp129s0
[DEBUG] LowCommandWriter - No motor command available!
Robot: G1-HG
Motors: 29
Message type: MessageType.HG
Control mode set to: PR
Current robot state: <unitree_interface.LowState object at 0x7ace11868470>
Successfully connected to the robot
Initialize Dex3_1_Controller...
🚀 Using new unified unitree_interface API
HandInterface initialized: LeftHand on interface: enp129s0 (re_init: false)
[DEBUG] HandCommandWriter - No hand command available for LeftHand
HandInterface initialized: RightHand on interface: enp129s0 (re_init: false)
✅ LeftHand initialized
✅ RightHand initialized
[Dex3_1_Controller] left_hand_state_array: [0. 0. 0. 0. 0. 0. 0.]
right_hand_state_array: [0. 0. 0. 0. 0. 0. 0.]
🔧 Initializing hands with default poses using new API...
Initialize Dex3_1_Controller OK!
2025-12-26 10:13:03.113795274 [E:onnxruntime:Default, provider_bridge_ort.cc:1992 TryGetProviderInfo_CUDA] /onnxruntime_src/onnxruntime/core/session/provider_bridge_ort.cc:1637 onnxruntime::Provider& onnxruntime::ProviderLibrary::Get() [ONNXRuntimeError] : 1 : FAIL : Failed to load library libonnxruntime_providers_cuda.so with error: libcudnn.so.9: cannot open shared object file: No such file or directory
2025-12-26 10:13:03.114154187 [W:onnxruntime:Default, onnxruntime_pybind_state.cc:965 CreateExecutionProviderInstance] Failed to create CUDAExecutionProvider. Require cuDNN 9.* and CUDA 12.*. Please install all dependencies as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.
ONNX policy loaded from /home/jqjs/workspace/TWIST2/assets/ckpts/twist2_1017_25k.onnx using providers: ['CPUExecutionProvider']
TWIST2 Real Controller Configuration:
n_mimic_obs: 35
n_proprio: 92
n_obs_single: 127
history_len: 10
total_obs_size: 1432
Body action smoothing enabled with alpha=0.5
Press START on remote to move to default position ...
Waiting for the start signal to move to default pos...
[DEBUG] LowCommandWriter - No motor command available!
[DEBUG] LowCommandWriter - No motor command available!
Moving to default pos.
Now in default position, press A to continue ...
Enter default pos state. Waiting for the Button A signal...
Robot will hold default pos. If needed, do other checks here.
Begin main TWIST2 policy loop. Press [Select] on remote to exit.
/home/jqjs/anaconda3/envs/twist2/lib/python3.8/site-packages/torch/cuda/init.py:218: UserWarning:
NVIDIA GeForce RTX 5090 Laptop GPU with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5090 Laptop GPU GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
warnings.warn(
另外这种情况是突然发生的，在这之前我已成功实现全身遥操，成功全身遥操的那一次，操作员动作过大，机器人摔倒过，但现在重启电脑和更换策略，这种情况依然发生