This paper formalizes the problem of delayed observation in multi-agent systems and proposes a solution framework, achieving superior performance on classic benchmarks with delayed observation.