Question about close-loop with LaneGCN

[zprhhs]

I trained 20 epochs with the LaneGCN provided by nuplan-devkit, and used the entire mini dataset. When doing closed-loop simulation on several scenes, I found that the ego-car always drives to the non-driveable area. Are there some problems with my experiments?

python nuplan/planning/script/run_training.py \
    experiment_name=vector_baseline_experiment \
    py_func=train \
    +training=training_vector_model \
    scenario_builder=nuplan_mini \
    lightning.trainer.params.max_epochs=20 \
    data_loader.params.batch_size=16 \
    data_loader.params.num_workers=8 

[JingyuQian]

Have you checked the tensorboard to see if the logged losses and sample images look normal? I believe the problem may arise from many sources.

Also, I'd suggest doing an open-loop simulation first, ensure the outcome predictions look normal, then proceed to closed-loop. That way you can tell if the problem comes from the model itself or the configuration of closed-loop controllers.

[zprhhs]

training log: https://ktbky.kaaass.net/paste//?074baee96c0fc3cb#A5SzLgb6snemowBHuEMg9G5vGNqhkWYNsgsTtzfqqtS6 open-loop vs close-loop: Compared with the results of open-loop and closed-loop (20 frames), I am not sure whether the problem is caused by the model. It seems that the results of open-loop seem OK, but sometimes it can not accurately use the lane information for planning(such as the second picture in the second row and the last picture in the second row). Do you have this problem when training LaneGCN?

[ShawWangWX]

Hi @zprhhs ,

Have you met the following error when running close loop simulation? If so, can you give me some instrutions?

''' QObject::moveToThread: Current thread (0x8b03900) is not the object's thread (0xa06e700). Cannot move to target thread (0x8b03900)

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/eigen/anaconda3/envs/nuplan/lib/python3.9/site-packages/cv2/qt/plugins" even though it was found. This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem. '''

[zprhhs]

Hi @zprhhs ,

Have you met the following error when running close loop simulation? If so, can you give me some instrutions?

''' QObject::moveToThread: Current thread (0x8b03900) is not the object's thread (0xa06e700). Cannot move to target thread (0x8b03900)

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/eigen/anaconda3/envs/nuplan/lib/python3.9/site-packages/cv2/qt/plugins" even though it was found. This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem. '''

sorry, I haven't encountered this issue during the testing process.

[liuandian]

Hi @zprhhs ,

Have you met the following error when running close loop simulation? If so, can you give me some instrutions?

''' QObject::moveToThread: Current thread (0x8b03900) is not the object's thread (0xa06e700). Cannot move to target thread (0x8b03900)

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in "/home/eigen/anaconda3/envs/nuplan/lib/python3.9/site-packages/cv2/qt/plugins" even though it was found. This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem. '''

have you solved this problem?