We propose a decentralized framework to solve a coordination problem for multi-agent systems consisting of heterogeneous agents, each of which uses only local information based on limited sensing and communication capabilities. In the proposed method, slower heavy-duty robots are each assigned a task specification specified in metric interval temporal logic. These specifications express complex, time-bounded tasks that are potentially dependent on other agents' actions. Heavy-duty robots update their task plans upon receiving a cooperative request from other heavy-duty robots in order to complete cooperative tasks. These requests are transmitted by the more agile light-duty robots responsible for information exchange, which systematically pursue heavy-duty robots. Our work in progress aims to present the framework together with a set of assumptions under which the solution is complete. We also aim to evaluate the framework on a series of use cases motivated by search and rescue.