one unsolved problem for multi agent dqn

use sum tree for priority experience replay

use curiosity for exploration

use multi(2) agent for solving problem

but not works...