MindMaker: Deep Reinforcement Learning (SingleNode)

Aaron Krumins - 6월 3, 2019

Design Self-learning NPCs using Deep Reinforcement Learning (A2C, PPO, TD3, ACER, DQN, SAC)

  • 지원 플랫폼
  • 지원 엔진 버전
    4.26
  • 다운로드 유형
    전체 프로젝트
    이 제품은 언리얼 엔진 프로젝트 전체 폴더와 함께 환경 설정 파일, 콘텐츠 파일 및 .uproject 파일을 포함하고 있으며 새로운 프로젝트를 생성할 때 템플릿으로 사용할 수 있습니다.

Video: https://youtu.be/rgQWJ5bkVkk

Add life like realism to video game characters using Deep Reinforcement Learning – a machine learning approach that combines neural networks with a learning model to sculpt agent behavior. Using the DRL Learning Engine, an AI can be made to perform any combination of actions involving objects in it's environment simply by changing a reward function.

With these tools, games and simulations can function as environments for training autonomous learning agents. Possible use cases include robotic simulation, autonomous driving, generative architecture, procedural graphics and much more. For game developers, the use cases for self-optimizing agents include controlling NPC behavior (in a variety of settings such as multi-agent and adversarial), prototyping game design decisions, and automated testing of game builds.

The MindMaker DRL Learning Engine *: A functioning version of the DRL Learning Engine is included with project. Algorithms presently supported include Stable Baselines : Actor Critic ( A2C ), Sample Efficient Actor-Critic with Experience Replay (ACER), Actor Critic using Kronecker-Factored Trust Region ( ACKTR ), Deep Q Network ( DQN ), Proximal Policy Optimization ( PPO ), Soft Actor Critic ( SAC ), Twin Delayed DDPG ( TD3 ). This is a single node version of the algorithms designed for use on a stand alone machine rather than a distributed collection of computers.

* The DRL Learning Engine is a standalone executable contained within the project files that is launched at the commence of play by the MindMaker AI Plugin

기술적 세부사항

Features: 

·        Deep Learning for Large Observation and Action Spaces

·        Modular Design to scale across multiple projects

·        Save and Load Previously Trained Behaviors

·        Support for Multiple Rewards For Complex Behaviors

·        Direct Access to Neural Network via Blueprints

·        Detailed Explanations and Tutorial

·        No Expertise with Neural Networks Required

· Includes the required MindMaker AI Plugin

Number of Blueprints: 7

Input: NA

Network Replicated: No

Supported Development Platforms: Win 64

Supported Target Build Platforms: Win 64

Documentation: https://www.autonomousduck.com/mindmaker.html

Tutorials and Examples:

Cart Pole: Creating a Custom Deep Reinforcement Learning Environment in UE4

Automated Stock Trading: Build a Bitcoin Bot In Unreal Engine 4

Further Resources:

Intro to Reinforcement Learning for Video Game AI