In this paper, we consider dynamic multirobot tasks that can be done by any of the robots, but only with the assistance of any other robot. We propose a novel approach based on the concept of 'assistance networks' with two complementary aspects, namely assistant finding and network topology update. Each robot, encountering a new task, seeks an assisting robot among its immediate neighbors in the assistance network in a decentralized manner. The network topology is defined based on pairwise stability via payoff functions that consider general task-related guidelines. As such, the number of potential assisting robots can be ensured a priori depending on tasks' requirements. As robots move around, the topology is updated via pairwise games. If the games are conducted by a network coordinator, each game is shown to result in a pairwise stable network. A series of simulation and experimental results in a variety of different scenarios demonstrate that the robots are able to get assistance or give assistance flexibly.