— This paper is about collision avoidance of crowd robots. For this purpose a model of potential field is proposed. This potential field, generated by a neural network, is unique to each robot. It changes in an intelligent way depending on the surroundings, basically on the space the robot disposes with. The potential field interacts with the surroundings, and calculates the next reference point the robot should move to. This paper extends with an advantageous property the research results of Mihoko Niitsuma about a movement model of crowd mobile robots in order to more efficiently avoid collisions between the robots. In her work the collision avoidance was solved with quasi Coulomb force. In her solution divergence occurred in some cases (e.g. the robot could not reach the destination point). The goal of our study is to avoid this divergence by replacing the quasi Coulomb field with a more flexible potential field solution.