AIDOG10

Overview
It is a program of Deep Reinforcement Learning using Unity.
Please use it for Deep Rinforcement Learning experiments.
A bipedal, humorous dog-shaped robot performs four tasks:

task 1: Sumou (Japanese wrestling)
task 2: Boxing
task 3: RUN: Running to any nearby location and staying in any direction
task 4: RUN2: Running in any direction

This package includes a TensorFlow2 sample script. It is a deep reinforcement learning script of TensorFlow2. Action policies of AI dog in this package are generated with it.

User's benefits

1. For students (high school, university and graduate students), scientists and researchers
Useful for deep reinforcement learning research.
Using this program, you can create log information of the movement of a biped robot.
That is, situation, action, next situation, reward. These are the information required for deep reinforcement learning.
This program receives the neural network parameters in text files.
Thus, you can use any deep learning library to learn action policies and pass parameters.

2. For game developers
Using the functions of Task 3 and Task 4, a biped walking robot can be moved to any location and stopped in any direction.
When creating an action game or sports game using this function, the natural movement of the character is generated by the action policy according to the situation, so you can create an interesting and unique game. People will be simply interested in the movement of the character.

Functions
1. Operation check of action policy learned by deep reinforcement learning
Executing learned action policies for the above four tasks
2. Creating log data needed to learn action policies
3. Testing learned action policies

Technical specifications
DirectX: DX11
.NET: .NET4.x
required open-source software: None

the number of polygons of 3D models
Dog: 17,640
Arrow: 120

the resolution of the textures
lawn: 1,200x1,200
lawn_small: 128x128
ring_small: 128x128
tawara2: 128x128
button: 64x64