Open hoonm1n opened 1 year ago
When i modified that line in srnn_model.py, another error occurred like under line.
Traceback (most recent call last):
File "/root/dsrnn_ws/CrowdNav_Prediction_AttnGraph/train.py", line 247, in
Do you want to add trajectory prediction? If not, change env_name
in arguments.py
to CrowdSimVarNum-v0
.
Otherwise, S-RNN is not compatible with prediction right now, so I'll have to fix it.
Does that mean that the old network's weights can't be trained in the current repository(CrowdNav_Prediction_AttnGraph) yet?
This repo has several different gym environments (see readme -> training -> ii). S-RNN can be trained with CrowdSimVarNum-v0
environment, which does not have human trajectory prediction. But it cannot be trained with other environments with prediction.
Hi @Shuijing725 !. I have a issue about your code in training with srnn. I changed robot.policy in config.py from "selfAttn_merge_srnn" to "srnn". Then when i run the train.py, the error occurred like under line.
`<Monitor<CrowdSimPredRealGST>>
No ghost version.
new gst
new st model
LOADED MODEL
device: cuda:0
Traceback (most recent call last): File "/root/dsrnn_ws/CrowdNav_Prediction_AttnGraph/train.py", line 247, in
main()
File "/root/dsrnn_ws/CrowdNav_Prediction_AttnGraph/train.py", line 90, in main
actor_critic = Policy(
File "/root/dsrnn_ws/CrowdNav_Prediction_AttnGraph/rl/networks/model.py", line 28, in init
self.base = base(obs_shape, base_kwargs)
File "/root/dsrnn_ws/CrowdNav_Prediction_AttnGraph/rl/networks/srnn_model.py", line 378, in init
robot_size = 7 if args.env_type == 'crowd_sim' else 2
AttributeError: 'Namespace' object has no attribute 'env_type'. Did you mean: 'env_name'`?
Could you solve this issue? Thank you for reading!