An overview of the development of modern approaches to conflict prevention between aircraft based on deep reinforcement learning is given. The basic concept of reinforcement learning and some fundamental algorithms used for aircraft conflict prevention are reviewed. Models with discrete and continuous actions for conflict prevention in two-dimensional and three-dimensional airspaces during flight along fixed trajectories or in free flight are presented. Various approaches to representing information about the state of the airspace (using a state vector and as a graph) and different types of interaction between aircraft (based on information about the state of surrounding aircraft or through message exchange) are considered.