Link to Github: github.com/berickcook/experiential-minecraft
Berick Cook
AIRIS plays Minecraft using SingularityNet's Vereya API.
Link to Github: github.com/berickcook/experiential-minecraft
Link to Github: github.com/berickcook/experiential-minecraft
updated 2 years ago
Link to Github: github.com/berickcook/experiential-minecraft
0:48 What is it and how does it work?
10:50 The test environment
11:52 AIRIS in action
22:48 The Features
28:02 Current limitations and future work
AIRIS is a causality based approach to Machine Learning
In this video I describe what the AIRIS Project is, how it works, show what it can do, and discuss what it can't do yet.
Project Features:
Small Data
Explainable
Transparent
Goal Agnostic
Continuous Learning
FAQ
Is there a paper?
I would love to have a paper about this, but I'm a self-taught hobbyist game developer not an academic. I've never written a paper and wouldn't know how. If you have experience in writing ML papers and can help, please let me know!
Where's the code?
I have not yet made the code for this version in this video available. However, there is an old version on GitHub called AIRIS_Public. Fair warning, the code is a mess...
Attributions
AIRIS
Author: Berick Cook
http://airis-ai.com
Infinitely wide neural network
Author: Tom Small
commons.wikimedia.org/wiki/File:Infinitely_wide_neural_network.webm
License: creativecommons.org/licenses/by-sa/4.0
The Rube Goldberg Idea Machine
Author: Carmen Angelillo
commons.wikimedia.org/wiki/File:The_Rube_Goldberg_Idea_Machine.webm
License: creativecommons.org/licenses/by/3.0
Blender3D BouncingSoftbodyBall
Author: SoylentGreen
commons.wikimedia.org/wiki/File:Blender3D_BouncingSoftbodyBall.ogv
License:http://creativecommons.org/licenses/by-sa/3.0
Atari Games
Author: Farama Foundation
gymnasium.farama.org/environments/atari
ProcGen Benchmark
Author: OpenAI
openai.com/research/procgen-benchmark
Franka Kitchen
Author: Farama Foundation
robotics.farama.org/envs/franka_kitchen
Song: Dreamer by Kevin MacLeod
Free download: filmmusic.io/song/3676-dreamer
License (CC BY 4.0): filmmusic.io/standard-license
Song: The Lift by Kevin MacLeod
Free download: filmmusic.io/song/5009-the-lift
License (CC BY 4.0): filmmusic.io/standard-license
Song: Chipper Doodle v2 by Kevin MacLeod
Free download: filmmusic.io/song/3500-chipper-doodle-v2
License (CC BY 4.0): filmmusic.io/standard-license
Song: IFeel by Lilo Sound
Free download: filmmusic.io/song/5871-ifeel
License (CC BY 4.0): filmmusic.io/standard-license
Song: Werq by Kevin MacLeod
Free download: filmmusic.io/song/4616-werq
License (CC BY 4.0): filmmusic.io/standard-license
This is a video of the initial "door" test. I had disabled AIRIS's ability to model objects that get collected (making them disappear from the model). I was surprised to see that it was still capable of finding the solution despite the disadvantage.
This is a set of test footage from the first autonomous tests of AIRIS. The first part is AIRIS playing the game with no prior knowledge of how the game works. It learns how to play the game just from raw pixel inputs. The second part is me playing the game while AIRIS watches and learns. The third part is AIRIS playing the game using my gameplay as a knowledge foundation.
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
"Chill Wave" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
This is AIRIS mastering the 3 level test. This test ran for 21 hours.
AIRIS has mastered the repetition test. This is the condensed version of a repetition test that ran for 14 hours (this test previously took 12 days, failed, and crashed). AIRIS is consistently able to reuse learned information to achieve its goals.
Previous AIRIS video: http://www.youtube.com/watch?v=QxzrMKqwaWM
AIRIS does 5 repetitions of 10 levels of the grid world puzzle game. This demonstrates AIRIS's ability to use both exploration and exploitation to not only achieve a given task, but to get better at achieving the task as it gains more experience.
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
"Chill Wave" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
This is a small demo of my unsupervised, model based, reinforcement learning system called AIRIS. It learns how to complete 6 levels of a grid-world based puzzle game with no prior training.
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
AIRIS demonstrates a powerful feature called "Knowledge Transfer". I created 10 new levels, trained Agent 1 to complete the first 5 levels and Agent 2 to complete the last 5 levels. I then transfer the knowledge of Agent 2 into Agent 1 to create Agent 3. Agent 3 then completes all 10 levels.
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
This is the second run of the fire puzzle room. AIRIS must collect the extinguisher, put out at least one of the fires to grab the key, unlock the door, and collect the battery to win.
I've implemented the ability for AIRIS to model the collection of items, but it's still a little buggy. Even so, AIRIS is able learn how to solve the room.
At 9:00, I directed AIRIS to only set its goal (the green highlights) to the battery to see whether or not it actually understood the tasks necessary to complete the goal and wasn't just completing the level by randomly completing the other goals. And while the buggy collection recognition did hinder it, it was successful!
This is the first video of the new AIRIS Prototype. This video demonstrates the observational capabilities of the system. Unlike previous versions of AIRIS, the Beta Prototype uses raw pixel data for inputs. This will allow for a much broader range of applications.
This is the first test of AIRIS experiencing a level change. There are 3 basic levels. Each level is completed when the batteries are collected. At the end of level 3, it restarts back at level 1.
This is footage of AIRIS attempting to learn how to complete a more complicated, multi-goal level. There are two new elements for it to contend with: Fire and an extinguisher. If it touches the fire before collecting the extinguisher, the level will restart. It has to collect the fire extinguisher, put out the fire by walking over it, collect the key, unlock the door, and collect the battery.
This is the extended version of the Repetition test. I let AIRIS run for 12 days straight to see what it would do.
This is a video of a test run of a new prediction algorithm. It works pretty well up until AIRIS goes insane (gets stuck in a loop of trying the same thing over and over).
This is condensed footage of an 18 hour long test run in which AIRIS is presented with a battery that continuously switches back and forth between 2 locations upon collection. This test was designed to see how well AIRIS is able to re-use previously learned information to complete a repetitive task.
This is condensed footage of an 8 hour long test run in which AIRIS completes the task of collecting the batteries intentionally for the first time.
This is a condensed version of a test run of my Artificial General Intelligence project called "AIRIS". The actions took place over a 20 hour long period and have been condensed into 1 frame per second. I manually captured each frame so there may be a frame or two missing here and there!
kongregate.com/games/terra0nova/aground
kongregate.com/games/terra0nova/aground
kongregate.com/games/terra0nova/aground
In this stream, I'll be showing off what AIRIS is and what it can do so far.
For more information visit: http://airis-ai.com
http://airis-ai.com
Music provided generously by Kevin MacLeod.
"Dreamer" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
"Chill Wave" Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 3.0 License
http://creativecommons.org/licenses/by/3.0
long game jam hosted on itch.io, I'll be playing as many of the entries as I
can manage!
http://glenaskey.itch.io/hoverhatsu
My entry:
http://berickcook.itch.io/lost-tomb
long game jam hosted on itch.io, I'll be playing as many of the entries as I
can manage!
http://paintedsky.itch.io/procedural-quest-generator
My entry:
long game jam hosted on itch.io, I'll be playing as many of the entries as I
can manage!
http://jgansaown.itch.io/procedural-shooter
My entry:
http://berickcook.itch.io/lost-tomb
http://orionzwiebel.itch.io/maze
My entry:
http://berickcook.itch.io/lost-tomb
http://gamejolt.com/games/dr-u-and-the-communist-space-monkeys/107389
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/don-t-just-stand-there/107386
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/docka/106592
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/do-or-die/107300
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/did-you-see-that/106818
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/depraved-hellbeasts/106728
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/deep-in-darkness/107200
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/death-diary/107402
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/darkest-night/107379
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/dark-life/107367
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/cut-away/106992
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/curse-of-theseus/107032
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/chain-letter/107264
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/cave-of-creatures/107345
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/bloodworm/107313
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/blind/107388
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/blade-s-darkness/106640
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/belloura/107085
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/who-s-there/107404
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/mrs-e-in-the-house-of-broken-dolls-asylumjam2015/106533
My entry:
http://gamejolt.com/games/lost-tomb/107403
http://gamejolt.com/games/through-the-valley-of-the-shadow/107369
My entry:
http://gamejolt.com/games/lost-tomb/107403


