• We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
...maybe 'not all that difficult' is relative.
Well, in the impossible or trivial sense of the word. It certainly wasn't easy to learn. :p
And isn't there something like free use of GEM engine for nonprofit games, like with Cry engine, unreal engine or Unity?
Couldn't say myself, but it might be worth looking into. (There is a similar thread over on the HoA dev forums, FWIW.)
 
As majesty 2 graphics has a very strong visual impact on the game, I believe that having a level editor that would work well with terrain tessellation, pass-ability (water, mountains, land), texture mapping, object placing etc would take the game 1 more year to develop at least. It is a feature for "lazy" people. I wish the level editor itself was a bit more flexible. Again, the level editor in m2 is very, very powerful. Nobody tried to learn it's greatness though.
 
In advance I'd like to say I read through a few pages, and skipped to the last page to sum up(there's a lot here).

I wanted to toss in a few ideas of my own from playing.

Heroes/Personality
-Set heroes personality that dictates stuff like armor color and style, to personality(which affects how they fight, spend, and what they take on in quests in addition to how level and class affect it).
-Heroes will occasionally take time off from adventuring in the tavern.
-When a hero is occasionally wandering, maybe he'll come on say too tough of a challenge for him alone so he's like holy shit lets get some more guys on this and will form parties on his own.
-Heroes will quest for artifacts based on their interests, and can trade them and gold to other heroes to help them on missions.

Better graphics are definitely a must here. I mean it doesn't have to be top of the line, but enough to push a medium end PC at top graphic settings.

(Noticed the last post was 5 months ago, hopefully a Majesty 3 is still in the stars and there's still interest.)
 
Noticed the last post was 5 months ago, hopefully a Majesty 3 is still in the stars and there's still interest.

Not sure there is, for the past couple of years, Inco and Paradox seemed to have focused their attention on Warlock, the TBS set in Ardania.
 
Some of those suggestions were more-or-less present in Maj1 (resting in taverns, class-based personalities affecting behavioural frequency, follow-and-support arrangements, seeking treasure/recovering items), but but they could certainly stand inclusion in a sequel, and hopefully further development. (Personally, I'm not concerned with graphics so much as I am with art style- Maj2 had a perfectly serviceable graphics engine, but I did not care at all for the design aesthetics.)

With that said, it's always nice to see the fans of this series have long memories... I think there's some hope in that.
 
  • 1
Reactions:
I hope there will be a third installment, bought the Majesty 2 collection on steam some time ago.
Just finished the expansion Monster Kingdom. Loved all the games. A lot more fun than Warlock ever was.
 
The key to making a great majesty game would be a involved AI engine. I had proposed neural networks trained on data mined from Heroes of Ardania a long, long time ago. This isn't the only solution, by far, but if they are going to invest in a new engine, might as well make a great game rather than merely a good one.
 
From what i've been reading, everyone wants a more advanced version of the AI from Maj1, more diverse constructions, the ability for heroes to form actual parties(instead of just wandering around together like in Maj1), heroes who actually get visual upgrades and can get upgraded in ranks.

Also, everyone here seems to be focused on only Maj1, Maj2, and other Simulation games. Personally, i think Paradox needs to take a look at the Stronghold series, as the AI for the villagers is pretty good most times, as well as the mentioned Total War series for its territory-to-territory stuff. if heroes would be wandering from the screen your kingdom is on, the devs would need to rethink the Majesty series's big map entirely. From Rangers, this would allow messages back to your kingdom about encroaching armies of monsters or heroes, a "War" mechanic per se.

For the suggestion about parties, both heroes only and henchmen led by heroes, what we would need would be a building similar to the Champion's Guild, where heroes with affinity with each other(say they accidently hunted the same target once) could meet and form a party to go and hunt down a high Attack Bounty target thats been sitting around for awhile. Of course, hiring henchmen would work similar.

As has been mentioned so far, the biggest pull of Majesty, and most simulation games about cities/kingdoms, is the people actually being able to "think" and act on their own. A virtuous "Warrior", is likely to go to an inn for a breather from his adventures, but if he sees a minotaur attacking a nearby hut, he should change targets to the enemy, then go have that meal he was planning on, or a drink. In other words, not only does the AI of the Heroes need to be able to change on the fly, but each hero also needs a Line Of Sight mechanic. Of course, the LOS mechanic would mean that unless a hero has direct access to the King or the king's advisor, they would have to check at a "adventurer's guild" type building to see if there is an Attack Bounty out across the map where the hero cant see.

As i read in a post, heroes being able to "evolve" would be an amazing new mechanic, especially when you bring ideas from other games into it. For example, a Warrior with high intelligence could evolve into a Spellsword with access to basic magics of various other groups, like Fireblast and a defensive spell. A Ranger who specializes in poisons, because he spent too much time in the Thieves's building, would become a Poison Archer. That was a bad example. However, if an evolving mechanic could be added, the strongest evolutions would have to be confirmed by the King(player). Say a Warrior managed to barely kill a dragon on his own, he could get to become either a Dragon Slayer, or a Champion, both of which would need to be acknowledged by the King. Other examples could be if a party went into a Goblin Fortress, but only 1 hero survived, they would gain the trait or Secondary/Sub class of "Survivor", and they're class would read as "Paladin(Survivor)".

Just plain having more options for heroes would be amazing. Examples could be, for non-religous heroes, from a Jousting Arena a Cavalier on horseback could be hired, or a Gladiator from a Colliseum type building where other heroes can go for combat training and betting on matches.

A way to bring back "Kingdom Decoration" type buildings would be that you could make "Kingdom Requests" at the "adventurer's guild" type building i mentioned earlier. By making a request to your heroes, they could go on a quest that takes them out of the map to collect relics. When the hero, or party, gets back, you would then be able to build a Monument to the achievement.

As was mentioned before, monsters need variety. A hero would get bored of the hero's life if all he is doing is slaughtering the same goblin over and over and over again. From what i remember about Maj1, the enemies with the biggest variety were the Goblins and Ratmen, which both had an almost kingdom-like variety of troops, and buildings when speaking of goblins. Dragons had what, 3 varieties and 1 was a boss? Kinda sad for them. Of course, if undead are all a single species, then they are comparable to Ratmen and Goblins.

The biggest change, to me, would be if you could start at, not a lvl.1 castle, nor a Outpost, but rather, the castle starting out as a simple Mayor's House, and from there leveling the "castle" up.

For connecting different missions together with the map idea from Maj1, you could make each "territory" persistant until you replay that map. Say you play a mission from Maj1's map, beat it, then move onto a mission next to the previous one. If you start to have a spot of trouble in the mission, you could call for aid from the settlement of the previous mission. The "Aid" sent could be high level heroes from the previous mission who have a timer before heading back, or supplies in the form of gold.

Speaking of missions, the unlocking mechanic from Maj1 was actually interesting. If they could rework that, bringing back the Map would be easy. Say you beat 2 missions and unlock the next, the new mission's intro would point out how the other mission's brought you to this one, making it into an actual story, rather then random stories across the map.
 
You're not the only one with a fondness for Stronghold- I'd love to see some elements from the Impressions citybuilder series as well- and I'd agree that the series might benefit from paying more attention to the large-scale strategic map. (e.g, where the magic sword you pick up in Tomb of the Dragon King goes on to be the key to slaying Vendral, or where Agrela's ring is needed to cure the plague from The Dark Forest, et cetera.)

Hero 'evolution' is a tricky point, because I wasn't fond of how it was handled in Maj2- e.g, where you select a hero, click a button, and then just 'upgrade' them on-the-spot like She-Ra or something. Partly because I'd prefer there were some kind of formal training/induction period, and partly because the hero themselves got no say in the matter. But I do like the notion of heroes being shaped (or scarred?) by their experiences.
 
BTW everyone, a recent development, Tensorflow API is becoming quite good and it has a C interface.
Since majesty 3 will probably not need cutting edge graphics, it would be possible to dedicate some GPU power for a neural network (learning) AI.
Real issue is probably anyone who's able to do this is probably working on something that pays better :3.
 
I have actually done a little work in tensorflow recently (part of my portfolio from a udacity course), using the python API. C++ seems to be de jure if you want to do work at a major games company though.

My heart seems to lie with abstract planning systems with maybe a little bit of NLP on the side, but I won't dismiss any particular approach out of hand. (It's something of an irony, because I was fascinated by neural nets when I was younger, and expert systems did nothing for me-n ow it's reversed, exactly when the industry is veering in the other direction. That's me, keeping pace with the times!)
 
I can do tensorflow and C++ if they are interested, but it would be an extremely bizarre use of my skills :) Reinforcement learning isn't related to my current work but it's not rocket science.
Neuralnets are particularly good for learning AI in gaming because it uses same resource that exist aplenty on beefy gaming PCs, as I pointed out above.
I don't really understand what you mean by abstract planning systems, if you mean expert systems, it's a lot of work. It also wouldn't feel organic in a game.
Btw, if someone does make reinforcement agent working in an RPG like environment (like majesty), it'd be a huge breakthrough for RPG genre in general, because all those NPCs in MMO can now suddenly be organic entities, the implication would be historical.
 
I actually read an article from the head of AI development at Unity recently, where he's hoping to roll out automated NPC AI as a standard feature within the industry, so I somewhat doubt I can get ahead of that curve at this point. The amazing thing is that it hasn't happened 10-15 years ago.

I suspect that fancy AI has had minimal value to the AAA industry before now- their own production values and attendant production costs have trapped them in the mold of producing games with linear narratives and low-to-middling difficulty curves, such that flexible, high-intelligence agents would have been wasted in that context. (That, and moravec's paradox ensuring that voice/speech/motion wouldn't have been able to compete with human actors.)


An example of what I was thinking you could do with an abstract planning system (like, say, STRIPS or it's descendants) might go as follows:
Code:
define travels(hero, place):
  require:
    borders(location(hero), place)
  effects:
    location(hero) = place

define close_to(a, b):
  return location(a) == location(b)

define pick_up(hero, item):
  require:
    close_to(hero, item)
  effects:
    has(hero, item)

define slay_dragon(hero, target):
  require:
    is_dragon(target)
    has(hero, magic_sword)
    close_to(hero, target)
    has_skill(hero, melee)
  effects:
    alive(vendral) = false
Code:
assert:
  borders(mayhew, mount_hellfire)
  borders(mayhew, tomb_of_dragon_king)
  location(magic_sword) = tomb_of_dragon_king
  location(sir_rodric) = mayhew
  location(vendral) = mount_hellfire
  is_dragon(vendral)
  has_skill(sir_rodric, melee)
  able(roger, travels, pick_up, slay_dragon)
  set_goal(sir_rodric, (alive(vendral) == false))

get_plan(sir_rodric) ->
  travels(sir_rodric, tomb_of_dragon_king)
  pick_up(sir_rodric, magic_sword)
  travels(sir_rodric, mayhew)
  travels(sir_rodric, mount_hellfire)
  slay_dragon(sir_rodric, vendral)

Basically similar to the monkey-and-banana problem, though there are lots of potential complications- ideally you'd want to handle probabilistic outcomes, loops and flow-control, fuzzy or numeric values, et cetera. I've been working on a prototype for handling this kind of logic- mostly as a self-education project, since I'm sure the state of the art is much fancier- but I might be able to demo something simple in a few weeks.
 
Last edited:
Mathematically, this is equivalent to neural net.
If you instance an input layer neuron with value equal to the output at a low level function (e.g. close_to). The rest of your code specify output of a static network.
The optimizer would end up outputting the results of get_plan through back propagation anyway :3 (in this case, it is the only solution)
In the cases with multiple solutions, however, the neural net might get stuck on a local minimum, which is more organic, abstract planning might lead to either no solution or best solution each time, unless you intentionally randomize it.

I think hardware powerful for reinforcement agents capable of playing Wow simply wasn't there for the last 15 years ><
 
I'm sure that a toy example like this can be translated into a series of inputs/outputs for a static network, but as the number of potential actions, agents, locations, etc. grow, then the space of permutations becomes large enough that I don't think you can cover it that way.

Of course, the search space for planning also grows exponentially, so you need good heuristics for guiding the search, but there are pretty high-performing techniques for doing that these days, or for bypassing brute-force search entirely. The other advantage of this technique is that you can ask the AI to explain itself (e.g, ask Sir Rodric why he left Mayhew in a hurry.)

I could be wrong- maybe there are ways to encode this behaviour in those new-fangled RNNs, at least for a finite selection of actions/agents/etc.- but I think you'd need to give me a code snippet to show me the architecture?
 
For games, the usual bottleneck is performance. Full fledged simulation is not feasible for simple behavior most of the time. For example, for a shooter game, it does not make sense to have very complex learning AI, usually because most actions boil down to a simple "hide/shoot/run away" sequence. Neural networks are too cumbersome, too difficult to test, and take too long to train and change, whereas a simple scripting based AI is very tweakable and flexible. And fast. Very, very fast.

Very little games need the actual neural-network level AI. Even smaller amount can afford it.

I can say this much:
- Whoever comes up with a NN solution that is deterministically tweakable, easy to work with, and runs under 2 ms - will change the gaming industry forever :)
 
I think it's possible something similar to a neural net could be used for devising rough strategies in broad types of situations- e.g, where your enemies are mostly ratmen, or a lot of your team are fatigued, or the terrain is mostly forest, et cetera. You wouldn't have to worry so much about explosive growth in the number of variables if you're just handling a finite set of attributes like that.

You could keep the network quite small that way, so training would be fairly quick- maybe even fast enough that it can adapt to enemy tactics on the fly. (e.g, the enemy priestesses are fielding vampires with magic mirror, so one learns by observation that fireball is less of a good idea there.) It would only give rough rules-of-thumb, of course, but... that's what heuristics are there for, if you need something to guide a planning-search.


I can't really comment much on FPS games, but I gather that tactical decisions of where to lay ambush, where to gather drops, how to cooperate with team-mates, use suppressive fire, etc.- can get respectably complicated. The standout example is apparently F.E.A.R, where the developers explicitly commented on how little progress has been made in game AI over the past decade. I think the problem here has to be one of economics and culture, rather than hardware bottlenecks.
 
Gaming industry evolves in short bursts, or mini-breakthroughs, usually hardware-associated. There is nothing really innovative that the gaming industry comes up with. For example, all graphics algorithms that we use today in games are usually based on ideas from 10-15 years ago. We could not do them then, but we can now. I believe this is very similar for AI - perhaps it would make sense to have a self-learning scoped NN for many occasions as you mentioned, we simply are not there yet. There are only a handful of games that tried NNs or quazi-NNs - what comes to mind is Forza and Black&White. In most cases its simply not worth an effort. Yet. Key word is Yet. Perhaps in the not so distant future this will change.
 
I remember that Creatures was a fair hit back in the day, even with hardware from twenty years ago, so I'm surprised that there hasn't been more development in that area. (AI for The Sims apparently peaked with Sims 2 as well.)

I'm still more enamoured of planning systems, but I guess we'll see what turns out down the road.