Recently, research on multi-agent environments has been continuously conducted based on the development of research on single-agent environments. This paper deals with the study of collaborative task design and learning for multiple robots to cooperate in an environment where multiple robots operate. Unlike a single robot environment, the multi-robot environment has a non-stationary characteristic due to the relationship between robot actions affecting each other. Such non-stationary problems are highly complex and related to collaboration tasks because they affect the performance degradation of learning and the convergence time of learning models. In this paper, we present three types of collaborative task problems: Pick-Push-Place, Collaborative Lift, and Handover. We also discuss the implementation of each task and the validation of learning outcomes.