It's a very tough task to follow the texts. Let me pick up just one example from Abstract. The author wrote "We construct decentralized multiagent system which behavior describes multijointed robot." It's still not so bad, but it might be better to be "We construct a decentralized multiagent system which describes (controls) behaviors of multi-joint robot." But the sentence which follows is chaotic. That is, "... and learn every agent locally using different proposed techniques of collective reinforcement learning and compare their efficiency." This is more like a jigsaw puzzle. Each part is appropriate but order is unpredictable.

The reviewer thinks the author has studied this topic quite well. But unfortunately, all the techniques described here are nowadays a textbook level. And the experiments described here are more like good practices for an undergraduate course. The author wrote "RL is a new ...," but a technique proposed in 1990's can be said to be new?"  

If this paper survives the review procedure, the author should proofread, or maybe better ask someone else to proofread. Typos,  cares-mistakes, and essential mistakes are ubiquitous --  Just an example but "Multi-joined" or "multi-jointed?" Writing is quite good, but still the author should learn how the formal scientific articles are to be like.

Thus, the paper is still very immature, but not a very bad one. The reviewer expects this paper is accepted and the author will make his best to refine it toward the camera-ready submission.