Skip to content

test on the train set, the explored area is less than one. #5

@zhen037

Description

@zhen037

Hello author, your works are truly excellent. However, in my practical experience, I have observed certain phenomena and would like to receive your response. I generated a training set containing 4000 samples based on the training images in your project, which was used for training the model. Then I tested the model on this training set. I found that in some data, the machine tended to get stuck at the corners. Personally, I suspect this is because the reward points are very close to the corners, but in fact they are located in another room area. So the machine got stuck at the corner and tried to pass through the impassable walls to obtain the reward, resulting in an exploration area smaller than one.

Do you have any good suggestions? Looking forward to your reply!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions