Autonomous and not omniscient bot programming

[Neige]

Hello fellow developer,

I participated in some bot programming contest, and I love it.
I am wondering if you already thought of a true multi agent (or maybe you already did it)?
By true I means autonomous and not omniscient.
I got this idea from a great video that I just saw (sorry for non french) : https://youtu.be/5CaVhGTG8eA

Thanks for all your hard work

Neige

[sprkrd]

If I understand what you’re saying, you mean implementing a multi-agent bot contest in which knowledge is not centralized, right? So each bot would be like a different program, disconnected from the rest. If a bot A has a piece of knowledge that is of interest to robot B, it has to communicate this piece of knowledge through some kind of channel, correct?

I’m afraid that the current platform does not allow to control several disconnected entities (please correct me if I’m wrong). 1 program = 1 bot. Even when you have under your control several agents (like in the Pacman game from Spring2020), the knowledge is centralized and the program simply sends one instruction per agent. It’s not possible to spread the logic among several separate entities. Of course, you could handicap yourself and pretend that your agents have to find some way to pass information among them, if that’s your preference.

That being said, there is at least one cooperative game: Code a la Mode. Each player controls one cook, and they have to collaborate to complete orders. So it’s multiagent, but each player controls one and just one agent.

[Neige]

@sprkrd

If I understand what you’re saying, you mean implementing a multi-agent bot contest in which knowledge is not centralized, right? So each bot would be like a different program, disconnected from the rest. If a bot A has a piece of knowledge that is of interest to robot B, it has to communicate this piece of knowledge through some kind of channel, correct?

Yes that’s exactly it. Robot can communicate only if they are in range.

@sprkrd

I’m afraid that the current platform does not allow to control several disconnected entities (please correct me if I’m wrong). 1 program = 1 bot. Even when you have under your control several agents (like in the Pacman game from Spring2020), the knowledge is centralized and the program simply sends one instruction per agent. It’s not possible to spread the logic among several separate entities. Of course, you could handicap yourself and pretend that your agents have to find some way to pass information among them, if that’s your preference.

I checked the SDK an I think almost everything is here to setup something like that.
According to the codingame sdk
We can start a multiplayer game. Add multiple agent with the same source code.
The only miss part is the communication between agent

[sprkrd]

Then maybe you can be the first one to create a multi-agent contest

[eulerscheZahl]

@Neige

Add multiple agent with the same source code.

But you won’t be able to get an arena match with multiple instances of the same bot. Each player takes part in the game exactly once. When you have an 8 player game (which is the maximum allowed by the SDK), it’s 8 different bots.
You can have the same bot multiple times in the IDE for testing, but that has no effect on rankings.

On a side-note, I saw such a concept on another site, where you (=your bot) could play breakthrough with fog of war and limited vision for each piece. But I can’t find int anymore, “breakthrough” is such a generic name.