This paper proposes a way to implement potential attraction/repulsion fields. We consider the implementation of a multi-agent system that moves toward a common target in unbounded space, using swarming algorithms based on Reynolds rules. Three swarm modeling approaches are presented, jointly implementing a finite algorithm that ensures collision avoidance with all available barrier types. The algorithms are described in more detail in our previous paper [1]. We propose the joint application of the investigated swarming algorithm and a predator avoidance model based on reinforcement learning to provide collision avoidance with dynamically occurring barriers. Thus, it is possible to temporarily evade the target to ensure the safety of the agents' movement. The algorithm is based on Q-learning, the result of which is an action function. We consider the behavior of a multi-agent system modeled using the proposed approaches in a bounded space—a polygon or polygon. In this case, in addition to the described interactions, the movement of a group of agents is influenced by repulsive forces from walls. There is a problem of compensation of repulsive and attractive potentials, accompanied by braking of agent or ignoring of walls when moving to the target. This problem is proposed to be solved using function-voxel models. The principle of movement of agents according to the local geometric characteristics stored in the represented graphical M-images of the simulated polygon is described. In the paper, the solution of problems for hazard avoidance using the approach of potential fields, which are expressed by voxel surfaces, is obtained. The advantages of using these models and the need for an algorithm for predator avoidance are highlighted.