In this paper, we propose a collision avoidance algorithm for two nonholonomic tracked mobile robots transporting a single object based on a function-allocation concept. In this algorithm, a leader robot is controlled manually. A follower robot estimates the desired trajectory of the leader along its own heading direction and generates the motion for transporting the object and for avoiding obstacles by using an omni-directional vision sensor. We experimentally implement the proposed algorithm in a tracked mobile robots system, and illustrate the validity of the proposed control algorithm.