Comment by xidong_wu

This paper optimizes a controller/policy which will be used to agent itself in the next round