I'm glad I found your channel, as I myself am seeking to try to make my own small tactical RPG. Helps me understand how I should approach tactical algorythms and the systems behind. :)
I'm glad you found me too :) I get it man, it's hard to make these games so I'm glad I can help as many people as I can. We need more good tactical RPG's :)
I'm a first-time indie game dev for fun, and I've been looking for this for my D&D based game! Brilliant use of your Scenario class to evaluate values in Q Learning type fashion. Instantly subscribed, good sir!
The idea of behavior selector is pretty interesting! I've just started on my game dev journey and it's assuring to hear a more experienced developer coming up with a similar approach. For my approach. instead of going through all possible actions and calculating some sort of points, I'm working on having different "behavioral patterns", and having different enemies adopt different pattern based on the state of battle. This also allow players to give commands to their units to speed up the flow of easy battles. Aggressive: If cannot attack, move towards closest enemy. If can attack, perform strongest attack. Defensive: Move towards closest cover. Protector: Move towards closest ally and take position between the attack path of closest attacker and defended unit.
It's a fantastic start and I had something similar myself when I started. For me, my problem was the game was too easy. All I had to do was keep my soft characters behind and they'd never be in danger. And if my tank did get close to death I would swap who's in front and the enemy would never finish them off. But I guess you're just going to have a lot more behavior patterns than you described right? Very interesting honestly Another issue is predictability, if players always know what their enemy is going to do it might get boring pretty quick. I still have this problem myself too. Good luck on your journey my friend, these games are hard to make but we need more of them :)
Yep, there's going to be a lot more behavior patterns - and patterns will change over the course of the battle - e.g. if their HP drops below certain level, they will change their pattern, or if a bunch of defensive characters gets clustered up, they will become aggressive. I suppose another way to deal with predictability is to add some randomness, though that might open up a whole new set of problems XD Thanks for the well wishes! Wishing all the best for your game too!
@@gamedevaki Awesome! You actually just gave me a flashback to the gambit system in FF XII. And I'm sure lots of games use similar approaches. I think it's a great idea
Great video! I have yet to find an other channel that talks about the dev that goes behind grid based games! Keep going! I think you landed on gold with these tutorials!
Sick video. Been putting doing this in my own Tactics Game off for a while but I finally need to do it hahaha. This is an awesome reference and really well explained, it's gonna be a massive help. Thanks!
Nice this is how brute-force chess engines work, they take a position and look into the future permutations with full knowledge of what both sides can do to determine how much score an attempted next move results in. Doing that in complex positions in such a complex games needs some shortcuts like alpha beta pruning, but the base algorithm "minimax" sounds the same as what you describe.
Great video! It's really interesting to hear how your implementing your AI. Good luck redoing everything now and congratulations on (almost) 1000 subscribers.
I worked as a developer on the tactics card game Duelyst and we used essentially the same system of assigning values of all the possible actions and choosing the best one.
Hi there. I am also into turn based rpg building. My idea for the AI was to create a system similar to Final Fantasy 12 gambit system. But the way XCOM devs did this is pretty interesting. Keep up the good work. You have my sub.
(underrated ff game btw). You can do both. Gambit system I think is basically a behaviour selector. You would still need to figure out how to do the thing while thinking about position. For example if the gambit was hit strongest character with X spell. You would still need to find the character and move into an optimal position to do the attack. It all depends on the game I think.
I just started working on a SRPG and this was really helpful. Do you reckon you could train your AI using Tensor Flow to make it really challenging? Thanks, and good luck with your project!
Thanks mate, glad it helped. I would argue that using machine learning of any kind for enemy ai is a bad idea from a game design perspective. I think measuring difficulty and crafting encounters would be very difficult.
Hello! I'm totally not an expert on the theme, so take anything further only as seriously as you wish. You really can't afford to simply evaluate the position you character gets into via some magic set of constants. Unless you are a freaking genius, your AI 'll be extremely predictable with such a realization, not to mention that it will often react to a situation in an inadequate way - just because you are very likely to have some hols in the way you evaluate the position. The obvious question at this point is "But why is XCom so good, than?". It is because its creators ARE freaking geniuses, not to mention that they are a big experienced team. (Also, there are some specific things about the game itself, that i won't get into right now) So, what to do? Well... as it often is the Math shell save us. Or some of it's close relatives, anyhow. The short answer is that we need to use Mote Carlo search over the field of possible moves to go as deep as we can in min-max turn sequence There are some problems with computer going brrr and floating away in the sky, though. However, there's this amazing video ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-sQSL9j7W7uA.html . It talks about implementing a similar approach for a strategic card game, while kinda solving some of the problems with system requirements. Sure hope I helped. If you (or anyone for that matter) wants to discuss this stuff, they should totally pm me - id be glad to talk about such themes. P.S. Not some kind of fraud that wants money, just really enthusiastic about AI for turn-based tactics. P.P.S. Sorry for -possible- bad English - I'm not a native
Wow, what a fantastic comment. Thank you so much and don't worry your English is excellent. I'm definitely not an expert in any of this. My idea to eventually solve my predictability problem was to, rather than just taking the best scenario, take a group of them. And then randomly select one using a weighted random (leaning towards the best). No idea what that would look like in-game though and it may still be too predictable. But I will definitely look into that video you linked. I understand why you're interested in this. The level of depth seems to be infinite depending on how far you want to go. From a game design perspective, I'm planning on having small quick battles so having a very smart AI may not be necessary, but it's way too early to tell.
Hello, at around 3:11 you show an excert of a talk about XCOM Enemy Unkown and i would love to watch the original video of the talk. I cant find it sadly enough. Could someone help me if they know it please
Here you go mate www.gdcvault.com/play/1018058/AI-Postmortems-Assassin-s-Creed it's a talk about multiple games so you'll need to skip to the xcom bit. But honestly the whole thing is really good
No worries, glad you enjoyed. Its in the description :) www.gdcvault.com/play/1018058/AI-Postmortems-Assassin-s-Creed. Its only a part of the whole talk. Although it's all pretty interesting. I think xcom is the second speaker
Check out Motion Canvas by Aarthificial (he's another game dev on RU-vid). I think Motion Canvas could increase the production quality of your explanations and make them pop even more!
Can you suggest a good strategy rpg for psp aside from Tactics Ogre, FFT and Disgaea? The console got so very very many strategy rpg and I can't choose which one to try.
Oh jeeze, I'd say those three are the top dogs. Maybe the Valkyria Chronicle games. They're a very different style of srpg but I've always enjoyed them. Jeanne D'Arc is another one I remember being OK.
This is a strange question, say I wanted to make a game where you right click to move around the isometric grid, without turns. Would you say my best bet is to actually not do it in isometric at all, and instead do it as a 2D game from the top down perspective, and redraw the sprites from that perspective. Basically this is my question: Is starcraft a 2D game, or an isometric 2D game? Question 2: Do you think that you would be able to do a tutorial on how to make units spawn, and follow a path in this isometric tutorial series you have? Like let's say you wanted to make an isometric tower defense game? Or again, would you just straight up reccommend doing 2D again?
Hey Carbon, not strange at all. The map type has nothing to do with controls you can do anything you want. Not sure if it's a tower defence or an RTS your making, my approach would be different for both. Starcraft is actually a 3d game with a fixed camera. If you used a 3d map you could use a nav mesh and it will handle most of the pathfinding for you. But I've never played with it myself. Yes I can definitely make a tutorial for you. Unfortunately I'm behind on 2 other videos I promised so I can't say when that will be
