Curriculum Learning implementation problem
Hi!@pengzhenghao Could you explain how you specifically implemented Curriculum Learning in your paper? Thanks!
Another question, why is the success rate of the tollgate and parkinglot scenarios you presented here so much different from what was mentioned in the paper?
Another question, why is the success rate of the tollgate and parkinglot scenarios you presented here so much different from what was mentioned in the paper?
The environment has changed a lot and many bug fixes were there.
So just using the latest metadrive and report the latest performance will be good in your paper.
Curriculum Learning
https://github.com/decisionforce/CoPO/blob/86975e924df9f7d0bee701955228e6855ae3f9ff/copo_code/copo/algo_ippo/ippo_cl.py#L22
I believe we dynamically change the number of the agents in the environment as training goes.