begin{tikzpicture}[node distance=1cm, auto,]
 %nodes
 node[punkt] (agent) {Agent};
 node[below=of agent] (dummy) {};
 node[left=2.0 of dummy] (state) {$X_t$}
    edge[pil, bend left=45] (agent.west);
 node[left=3.0 of dummy] (reward) {$R_t$}
    edge[pil, bend left=45] (agent.west);
 node[punkt, inner sep=5pt,below=1.0cm of dummy]
    (environment) {Environment}
    edge[pil, bend left=45] (state.south)
    edge[pil, bend left=45] (reward.south);
 node[right=2.0cm of dummy] (action) {$A_t$}
    edge[pil,bend left=45] (agent.east)
    edge[pil, bend left=45] (environment.east);
 end{tikzpicture}