This paper focuses on decentralized coordination for small or medium groups of heterogeneous mobile robots with relatively low computational resources. Specifically, we consider coordinated obstacle avoidance techniques for mobile platforms performing high level tasks, such as patrolling or exploration. In more details, we propose the use of a greedy kinodynamic collision avoidance approach for the single robots and the use of the the Max-sum algorithm for multi-robot coordination. The system implementation and its testing are based on the popular robot middleware ROS and the gazebo simulation environment. Obtained results show that our distributed collision avoidance approach is able to achieve safe navigation in real-time with a very low overhead in terms of computation and communication.