Closed zhiyuan8 closed 4 months ago
As shown in AppAgent
https://github.com/mnotgod96/AppAgent/blob/main/scripts/task_executor.py#L204-L206
they use last_act in their prompt, which makes it easier to detect if there is a dead loop and we need to find another solution.
Also, they use explore phase to improve their task execution phase.
Could those improve accuracy?
Thanks for your suggestion. We have preserved the operation history, which is provided to the Mobile-Agent in each round. The operation history includes every action generated by the Mobile-Agent and the corresponding screenshots. It is stored in the format of a dialogue and has been simplified to enable the Mobile-Agent to recognize dead loops or erroneous operations. In our paper, we have showcased the performance of the Mobile-Agent when confronted with invalid operations.
As shown in AppAgent
https://github.com/mnotgod96/AppAgent/blob/main/scripts/task_executor.py#L204-L206
they use last_act in their prompt, which makes it easier to detect if there is a dead loop and we need to find another solution.
Also, they use explore phase to improve their task execution phase.
Could those improve accuracy?