OpenRA

.

.

See https://docs.openra.net/en/latest/playt ... rbotmodule and the following. Those traits are enabled with the bot type as condition. The result is the current "modular bot" architecture. You can replace any of the modules in mod code which makes it kinda flexible. The architecture is still very monolithic and you will run into trouble when multiple bot controller modules fight over the control of a unit.

https://github.com/OpenHV/OpenHV/blob/m ... /bots.yaml
https://github.com/Dzierzan/OpenSA/blob ... es/ai.yaml

has some examples of slightly customized AI bots.

Just read this and an idea popped up in my head. Any chances that someone will implement neuron-network based AI in OpenRA? There are several RTS Games, where OpenAI has learned to be tougher than human by analyzing thousands of replays. This AI wins most of the games, even against PRO-players. These kind of bots would be very good for OpenRa, especially in terms of training.

.

.

.

.

.

.

Yup, these ideas are very interesting, but I think that mission (levels, scenarios) are not as complicated in terms of AI. I mean, you still can make this with a simple code, without using an actual AI.

I am interested in architecture of OpenRA bots. How does the AI actually work?

I am not sure about this, but I suppose that when a human starts game versus bots, there is only one computer-AI, which controls an enemy "player" (buildings, troops, special abilities etc). I also suppose, that the bot has maphack (computer has no fog of war). If such statement is correct, than we face the real problem — AI knows environment and acts of human without actual investigation by troops. I did not see such behaviour in-game, but I think that AI only imitates fog of war by algorithm, and a computer-player have maphack. Thus, it is a cheat. Correct me, if I am wrong. One possible resolution is to add more AI's in-game, which will be separated. For example, Fog of War is implemented and controlled by one algorithm, and the actions of PC-player — by another. Algorithms are separated, so the computer player can not see what human does without investigation. Alien Isolation has such concept, check it out.

I am also interested about AI on the map itself. For example, there are civilians on some maps, or neutral tanks, buildings. Same with badgers. Which AI controls these? The same, which controls an AI-player, right?

Ronald, the thoughts about Wizard and trainings are very interesting. The idea of "undefeatable" AI, in my opinion, works well. If an artificial intelligence is decently human-alike, than you will have no problems by using it during training-session. The problem is, that the current AI may have "cheats". It builds faster, it produces units faster, AI may have constant maphack etc. Every human player, even from the master-league, will sometimes (at least several times during a game) forget (for a brief moment) to build something, or there will be kind of delay. Also human players can't have as good macro- and micro-, as AI (in theory). Especially micro. We are also limited in our devices. We can see only part of the map, there is a delay before we make a decision and use keyboard or mouse. Current AI can imitate such delays, but it is not fair anyway.

Everything below is sarcasm:

Actually, we already have an AI, which can win everyone. This AI also can start to learn from scratch, by analyzing replays. I am going to use it during the next tournament — I downloaded gigabytes of OpenRa replays, so now I have games of almost every player in the game. I will start a new iteration of AI before every game versus my opponent. The AI will learn patterns of behavior for each player, thus I can win everyone. The only thing i need is to spend several days before the match playing versus AI-copy. I will win next tournaments and am going to take all the cash////

.

.

Here's an interesting article about ai development and using it - https://www.cleveroad.com/blog/ai-in-ed ... -can-gain-. It's definitely worth reading

.