The Impact of a Set of Environmental Observations in the Problem of Acquiring Movement Skills in Three-Dimensional Space Using Reinforcement Learning Algorithms

In this paper, we study the influence of the composition of a set of environmental observations on the solution of the problem of acquiring movement skills by an agent in three-dimensional space. We found that some redundant data can slow down the learning process and interfere with problem solving. Experiments on the effects of rules are conducted in the Unity game engine's environment using the ML-Agents package. The algorithm under study, Soft Actor-Critic, was chosen because it is one of the best ways to learn how to move in three-dimensional space.