Comments on: Enjoy AI in Minecraft – Project Malmo and MineRL/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/Professional Development, Data ScienceWed, 08 Nov 2023 07:58:17 +0000hourly1http://wordpress.com/By: Wayne Mountan/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-42372Thu, 07 Sep 2023 23:41:03 +0000/?p=12437#comment-42372It’s pretty wild to think of an Ai playing Minecraft

Like

]]>
By: Run Reinforcement Learning on Ray Cluster (Microsoft Azure) – tsmatz/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-39304Fri, 08 Oct 2021 13:19:57 +0000/?p=12437#comment-39304[…] Project Malmo Tutorial (Post “Enjoy AI in Minecraft”) /2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/ […]

Like

]]>
By: Tsuyoshi Matsuzaki/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-39205Fri, 17 Sep 2021 21:28:20 +0000/?p=12437#comment-39205In reply to Freek.

Hi Freek-san. Please specify “ObservationFromFullStats” in mission and determine rewards in your custom env. You can get the following information (positions, life, etc) with this setting, “ObservationFromFullStats”.
https://microsoft.github.io/malmo/0.21.0/Schemas/MissionHandlers.html
If it doesn’t meet your requirements, please see other observation options you can use in the above document. (You can use other observation options together with ObservationFromFullStats.)

Like

]]>
By: Freek/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-39204Fri, 17 Sep 2021 18:44:36 +0000/?p=12437#comment-39204Would it be possible to extract inputs (like point of view) from Minecraft using the Malmo or MineRL libraries to determine the reward in Python itself? i.e. make a mission that doesn’t specify any rewards in the XML file but determines these in the Python code exclusively instead.

Like

]]>
By: Tsuyoshi Matsuzaki/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-35335Fri, 30 Oct 2020 02:01:27 +0000/?p=12437#comment-35335In reply to Tsuyoshi Matsuzaki.

Yes, the agent interacts with minecraft server (w/ malmo mod) using socket (TCP) communication.
I’m sorry since I’m not familiar with implementation details, but the agent (see AgentHost.cpp below) holds the socket communication as observations_server, video_server, etc. These “server” objects are all implemented as StringServer (StringServer.cpp), which uses TCPServer (TCPServer.cpp) for underlying TCP connection. As you can see in TCPServer.cpp, it handles the received message in its callback. When you peek states, these message will be returned to the client.
As you know, python implementation is just wrapping these native C++ implementation.
MineRL will also be using this malmo implementation at the bottom.

AgentHost.cpp
https://github.com/microsoft/malmo/blob/master/Malmo/src/AgentHost.cpp

StringServer.cpp
https://github.com/microsoft/malmo/blob/master/Malmo/src/StringServer.cpp

TCPServer.cpp
https://github.com/microsoft/malmo/blob/master/Malmo/src/TCPServer.cpp

Like

]]>
By: yang9004/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-35323Thu, 29 Oct 2020 09:08:36 +0000/?p=12437#comment-35323In reply to Tsuyoshi Matsuzaki.

Yes, I already change my code to peekWorldState()

But, I still want to know how can I obtain the world_state ( = agent_host.peekWorldState()) information in MineRL or Malmo. Do you know how to generate my own environment? I have read the code of MineRL and MalmoEnv , the key code seems talks about socket sendMessage to Minecraft Server or receive Message from Minecraft server, As far as I consider , the Minecraft server provide all kinds of message, but I can not find how to receive message including world_state, Do you have experience of this problem ? can You help me?

Like

]]>
By: Tsuyoshi Matsuzaki/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-35182Thu, 22 Oct 2020 01:46:23 +0000/?p=12437#comment-35182In reply to yang9004.

I think that more appropriate function for getting world state might be AgentHost::peekWorldState() in malmo. (See https://github.com/microsoft/malmo/pull/96 )
peekWorldState() can also be used in python with MalmoPython (see the source code in https://github.com/microsoft/malmo/tree/master/Malmo/samples/Python_examples ), but I think you might have to generate your own environments, when you need gym-like environments with peekWorldState().

import MalmoPython
agent_host = MalmoPython.AgentHost()
world_state = agent_host.peekWorldState()

Like

]]>
By: yang9004/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-35098Mon, 19 Oct 2020 13:39:05 +0000/?p=12437#comment-35098hi! Nice post !
I have a question that what if I want to use the world_state information (which is the important object in Original Malmo Python interface including all kinds of observations like observations from grid , observations from ray ) in Malmo or MineRL package, what should I do? Should I modify the package code of Malmo or MineRL ? As far as I consider the Malmo or MineRL just a python wrapper of Malmo server and interface.

Looking for your reply!
Great post !

Like

]]>
By: 360digitmgvizag/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-33736Thu, 27 Aug 2020 09:52:03 +0000/?p=12437#comment-33736Nice and helpful article!! Thanks for sharing such an informative post.

360DigiTMG

Like

]]>
By: Azure Update (2020.07.16) | ブチザッキ/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/comment-page-1/#comment-32973Wed, 15 Jul 2020 18:45:24 +0000/?p=12437#comment-32973[…] Enjoy AI in Minecraft (Malmo and MineRL) […]

Like

]]>