This paper presents a broadcast control framework for multi-agent coordination. Here, the "broadcast control" means governing multi-agent systems by sending the same signal to the all agents indiscriminatingly, under the assumption that there is no agent-to-agent communication. The problem considered here is to design the information to be broadcasted and the local actions of the agents, to achieve a given motion-coordination task. For solving this, it is first shown that the local actions have to include randomness for realizing the broadcast control. Based on this, we propose a broadcast control method, where the achievement degree of the task is broadcasted and the agents alternately perform random walk and deterministic walk. The proposed method is proven to asymptotically achieve the task with probability 1. The performance is demonstrated by numerical simulation and experiment. (C) 2013 Elsevier Ltd. All rights reserved.