Credit assignment in heterogeneous multi-agent reinforcement learning for fully cooperative tasks