counting

For an agent-based approach, or just multi-agent system, you don't actually need a complex structure, or even full profiling. The adaptive ability only requires certain internal (private) variable(s) for each agent, in evolutionary programming term, like a software gene. You can randomize it, mutate it, swap it, change it, etc. The base code/gene will then "present" itself and grow into different agents. Agent also doesn't need to be at individual level, it can be a house, a cart, a grid, a group of the same age bracket, even a whole sector of concept, and all the combination you can think of. The keys are letting agents to have enough freedom (autonomous), each agent is bounded by its local variables, and interact in parallel cooperatively. Using Pharaoh/Caesar series as examples, an agent isn't just a walker, its an entity of its building, the walker, the associated services/goods, and it's location, connectivity to the road. Each agent has its "local variables" (profile if you like) stored as its current labor level, its current placement, and the type of building it is, etc. They interact with each other via walker travelling, any change of the road system, walker will adapt to the change without player's direct control (degree of freedom). Generally in agent based model in academic research we let the emergence behavior evolve on their own, with some randomize process to find the right combination. But used in game design, it's players who are doing the trial and error and trying for different permutation arrangements (like human randomizes). Sometime players success and get interesting builds, other times they fail catastrophically with all the "FUN".

Impression's ancient city builder games, from Caesar IIi, Pharaoh, Zeus, to Emperor all have the basic walker-type agent mechanism. Strictly speaking it's also use a type of agent-based approach, since each unit of goods(cart pushers), walkers, have their own behavior logic and rules. However, it still uses some statistical approach to handle certain aspect of the city simulation, like population/census. Generally speaking, city building games rarely built with pure agent-based approach, otherwise it consumes too much computation resource (pure agent based simulation is mostly for academic studies, or "peculiar games" like Dwarf Fortress ). Also, Agent-based approach alone is not enough if you want some REALLY interesting results. In order to make agents interact with the "environment", and actually shape their "surroundings", you need to use adaptive strategies or learning mechanism on agents, so they can genuinely exhibit complex behaviors (A lot of the games only apply "dumb agents", who's behaviors are predictable regardless when and where with limited conditions, like walkers in Impression's city builder games if you know the rules well, even agents in Simcity 2013) ========================================== BTW Grid-like system is easy to implement with city simulation, even Simcity series themselves use it (until Simcity 4) with great success, and coordinate system is still fine-grid, albeit very tiny sparse grid. Also this has nothing to do with agent-based approach.

This works great on tablet. Do you guys have a time table of releasing official tablet-version? And can we early android tester get official release after May 15th? P.S. I didn't receive your official forum register confirm email, so I have to ask you guys here. And I can not find your board administrator email.

On a side note, since I'm familiar in the field of machine learning, I always thought it's strange that not many games are implemented with actual machine learning mechanism. As most players are aware of and usually frustrated with "dumb" AI behaviors, even in detail to-the-teeth game like dwarf fortress. I think part of your concerns can be resolved with learning mechanism. In the past, machine learning are not popular for game designers because it's resource heavy, but now it should be more favorable. An agent behave "dumb" is mostly caused by limited rules can be implemented with unique profiles (if you going millions agents it's also about time complexity), but if rules are adaptive, then you can more or less mitigate problems like volatility, or re-playability Also building budget should not be just about "a single resource" - "money", but actual building materials. You can have a evolved economic system 10 times the price but cost the same amount of materials, it's the relative ratio to other goods that effect the whole system balance (and player's conception of difficulty can be taught with proper tutorials, let players feel the effect of adaptive system gradually). After all it's bricks make buildings not "digits".

Emerging property is something that's interesting since you usually can't predict what may come out the other end. But certainly a good surprise if you don't always follow the realistic approach. (like dwarf fortress type, obviously emerging property comes out to be "FUN"). One thing about MUD though is that it has a fixed set of well established commands agents/players can issue and taking effect at a tick, hence we were forced to simulate details only at a certain level, and got a reasonable balance between gameplay and dynamics. I remember originally we wanted to appeal new players and provided cheap houses, but the approach of generating blank housing became a disaster. And housing market is just a predefined money sink. Later a player invented "rentals" when he wrote a script that sell and buy back floors at cheap price in regular intervals dynamically according to "his survey to new comers" (just raise and decrease opened floor level when players post requests on trading boards), so we modified it and implemented into NPC scripts thus generate a whole sector of NPC "rental markets" and created live real estate market, worked a lot better than MUD build-in inn keeper codes. Interestingly we have to write new sections of scripts - "not to travel" - for NPC, or they would flood cheap rent cities. (and someone later suggested being part of mechanism for travelling and tourism, which also worked nicely with more modifications). This kind of surprises are mostly fun but really need constant interaction between player community and developers. (It isn't a problem in MUD type MMORPG games, but might be a challenge for a simcity type) So a good way of approaching multiplayer scenario, might be the one you mentioned earlier that a player is acted only on behave of a single agent, has limited amount of influence to a group of agents, thus giving space for NPC agents generate their own emerging property and keep the illusion of a background world "breathes on its own". The real challenge will be to which level you let modding affect the system and which part of the micro behaviors default available for players that they feel their actions make a visible difference. (the true appeal to city building game experience IMO) P.S. local traders have competitive advantages over farther traders doesn't sound right to me. Local over global might be stable, but stagnate. I think it's the issue of coincidence of want, and goods utility should be the real driving force. After all "price" is just the ratio between money and goods. Money is a type of special commodity that need to be internal rather than external. In order to generate macro-level system, these agent should be simple enough only adapt steps at a time rather than calculated equilibrium point. Supply/demand generated from not so coherent aggregated requests, and price is just a reference, the real focus is the flow of useful goods. The base level needs to be solid before upper level can behave. An economic system can function without referencing "money" at all in a bargain system. I've read papers in experimental economics that uses agent-based approach and able to produce the emerging of "commodity money" with just basic trading-posts model.

I may share some insights and real experience of multiplayers working together to build cities in an economic simulation game. About a decade ago, we use MUD (text based MMO) engine/library and create a semi-graphical city building / economy simulation MUD. We reach about hundreds of simultaneous player logins and thousands of continuous players. A player like any other MMO can login as a character and move around the simulated world. But instead of mostly killing monsters, players mainly participate economic activities, from raw material productions (fishing/mining/farming/etc), buy/sell lands (required permits from real player mayors), building facilities/infrastructures/factories with materials from raw materials, start companies/corporations and building commercial buildings as retail stores and sell products (to NPC/players alike). Even playing the financial market, stocks, foreign exchange, or auction houses for inter-city trade. The interesting thing is that since it's using MUD engine, we can simply change the NPC/monsters script to make them act like economic agents. You can hire NPC as employees, and they in turn will use salaries to buy products they "need". And here is the fun part, you can also hire real players as well and so on. Players are essentially given same commands as NPC. So in the end, we have a hybrid society, where game wizards are policy makers and law enforcement agents who set up the rules for players to follow, mostly just for keeping trolls out (like constitution), and leave "civil laws" for players to decide and follow (we have cities essentially free to do anything, to cities tightly regulated to the teeth). Senior players (since running a city often require massive capital) are mayors and "local government" officers who keep their set of local laws within their boarders, cities/players can form alliances hence usually they are like a guild. Seasonal players who just like having fun, usually are corporate/institute owners and upper-class citizens who can join or leave a city (per request or at will depend on local laws). And last, new payers along with most NPCs are the middle class and labors who are hired by local governments or other corporations and gaining experiences/capitals. Most works are done "automatically". Since it's easy to write bots to perform script acts, so the line between NPC/players are not that clear. In fact with good enough scripts, NPC can be governors or officers, corporate owners, and we can adapt an NPC profile to a player profile at will, since the only difference between them is executing scripts locally using server resources or remotely with player computers. We even compete bot scripts with each other to test if a good enough AI program can function well and find good locations, determine prices, etc. Most of the time, mayors are player bots, since tax related task, or accepting applications are just too boring and usually accepted automatically with scripts. In-game script system allow players to perform some predefined tasks on schedules, essentially make players semi-NPC during offline hours. but most seasonal players prefer their own complex bots and execute remotely, essentially become zombies when they are not in front of their computers. A player can even use their vehicles (we have automobile industry simulated), and go site seeing. Not because there are pretty ANSI graphics to see, but playing as if it's a normal MUD (dwarf fortress adventure mod so to speak), where you venture into "rooms/buildings" created by players. Most of the time it's fun to see people typing humorous descriptions, even join player-made quests for rewards (Mayors often use quests to attract new players). Although the fun parts are still planning your businesses and running cities, express your own presence in the world, scripts are just mechanisms to help with boring repetitive tasks due to the nature of MUD engine we adopted. About the simulation size, due to server resource constrain, the over-all city wide unique NPCs are around thousands (globally in the entire world close to a hundred thousands) load their script according to their current roles. They have a set of predefined scripts activated in difference situations/timing. While at work they don't calculate prices, just repeating tasks every so often required by system. Off work, they will go "shopping" and find potential shops and the rest 1/3 of the time they are idle in their home (if they can afforded the rent) like they are asleep. Also we "copied" NPC behavior and added noises to their final decision, thus create a group of similar NPCs who travel together, and lived nearby, hence multiplying thousands of NPC (with their own unique profiles) into millions represented in game without the need of much additional server resources. We even at one point consider open MUD protocols (like APIs) for players' clients background services and help executing NPC scripts remotely. IMO, the key is to have enough details dwarf fortress style, thus able to create the emergent property that work organically, instead of by design. It's the modern equivalent to open-world games, and thankfully we don't focus on fancy graphics hence able to provide a fine multiplayer city building experience and detail agent-based simulation. Sadly it's also due to lack of graphics hence not able to attract enough new generation players.(we do have a project hope to update it with 3D graphic GUI, but it cost too much and still a work in progress). Although, the text based 2D MUD still have dozens of die-hard players online today. (Another factor is translation, we are stuck with Chinese contents generated mostly by locals, very small niche market). This bottom-up approach is quite interesting and a lot of fun when you build the foundation right.

This reminded me a lot of an old project/game I was with over a long time ago. We used MUD (does anyone still remember what it is?) library to create an online city simulation MUD - RevivalWorld (sorry it's Chinese-only official site, we are all locals). It even has real-time cities skyview map (players can develop multi level buildings to increase capacity and ofc it looks cool and give a great sense of accomplishment), complete with color codes. A sample of a particular city looked like this : Colors and regions like dark green forest, black roads, light yellow high ways, red bridges over rivers, brown mineral deposits, different kinds of industries and facilities with pink, light green, light blue color, etc. Even government facilities as purple, and light grey as commercials, and darker grey as residences. (This particular city is very well developed with lots residences, commercials, and low industries). It's updated in real time when there are changes and hourly. I think it's similar to your multiplayers combined with city building/simulation idea. I can make translation to English if you are interested in its mechanism. (even some library source code)