A Raspberry Pi and DeepLearning4J Brain That Plays Minecraft
summary
A functioning Java-based brain runs on a Raspberry Pi cluster using DeepLearning4J and RL4J with Project Malmo as the Minecraft interface and Zulu embedded JVM for ARM. Architecture maps a simple neocortex to four Pi Zero layers (plus a Pi 3 controller), exchanging state vectors about nearby blocks and using reinforcement learning / DQN-style policies to choose actions. Demo environment places random lava; agent learns to avoid death sometimes and sometimes dies — work in progress but runs end-to-end on cheap Pi hardware.
AI-written summary · about
chapters (23)
- 0:00Intro
- 1:32Inspiration
- 4:53Hierarchical Temporal Memory
- 6:27Introduction To Machine Learning
- 12:20Deep Learning Basics
- 15:58Neural Network Elements: Node
- 17:01Neural Network Elements: Layers
- 19:21Markov Decision Process
- 25:56Raspberry Pi Variations
- 26:40Pi Brain Architecture
- 27:06Raspberry Pi Cluster (Mk 2)
- 28:09Cluster Hat
- 29:02Why Java?
- 31:05Azul Zulu Java For Embedded
- 32:03Deep Learning4J (DL4J)
- 34:02DL4J Core Classes
- 34:56RL4J Basics
- 37:52Project Malmo
- 39:01Malmo Example Code (1)
- 41:25Demo Software Architecture
- 43:58Applying ML To Minecraft
- 45:24Conclusions
- 46:13Future Directions
related talks
49mA Raspberry Pi and DeepLearning4J Brain That Plays Minecraft
01h 09mMinecraft as AI Playground and Laboratory
41mReinforcement Learning and Minecraft
44mBuilding Smart Devices using Java on the RaspberryPi-An intro to Pi4J.
50mIEEE CoG 2019 - Industry Day - Katja Hofmann (Microsoft Research)
32mAI and Minecraft
46mRaspberry Pi with Java 9
50mApplying Machine Learning to IoT Use Cases with Deeplearning4j