Thank you for your interesting work.
In Object Relocation, when training two reward classifiers, both success and failed transitions need to be provided. How this failed transition is defined? Was it a sudden mistake during the transition collection process that led to the transition being judged as a failure? Or when only failed transitions are recorded, how should they be directly demonstrated as failed transitions (such as only picking up without moving, or moving without picking up)?
Because I saw that these two trained reward classifiers needed to be added in record_demo.py, but when running record_demo.py in Object Relocation, it was not very good to determine whether the task was successful or failed.
Looking forward to your insights.
Many thanks :)
Thank you for your interesting work.
In Object Relocation, when training two reward classifiers, both success and failed transitions need to be provided. How this failed transition is defined? Was it a sudden mistake during the transition collection process that led to the transition being judged as a failure? Or when only failed transitions are recorded, how should they be directly demonstrated as failed transitions (such as only picking up without moving, or moving without picking up)?
Because I saw that these two trained reward classifiers needed to be added in record_demo.py, but when running record_demo.py in Object Relocation, it was not very good to determine whether the task was successful or failed.
Looking forward to your insights.
Many thanks :)