Description

This laboratory allows you to learn and consolidate the basics of working with the TensorFlow machine learning platform on virtual sandbox sites. This complex works on the principle of reinforcement learning. The software package provides an interactive environment, including design and configuration of neural network parameters, programming the behavior of agents and the environment, visualization of the learning process, collection of experimental data and simulation of the task. The educational complex consists of the following modules:
• 3D engine for visualizing gameplay and simulating physical interaction
• Neural network model configuration module
• TensorFlow Sharp module for model execution
• Module for outputting results and visualizing linear regression
Within the framework of this software package, the following set of gaming problems is solved using a neural network and unsupervised learning:
• Simulation of an agent hitting a hoop and finding the optimal throwing force F at a given distance to the target X.
• Simulation of the passage of a two-dimensional route by an agent, with the condition of the absence of collisions at a given speed.
• Modeling the solution to the problem of parking an agent “car” into a parking space of a given size and position, with variable values ​​of the agent’s speed and turn.
PC included.