AI method Request

Your suggested scouting method is interesting - I can definitely see how it would be extremely efficient in certain circumstances, but it doesn't seem particularly adaptive and it also doesn't apply well to rectangular maps...

It's not particularly adaptive (not without major modification, anyway) because if I manually send out a scout in a straight line off in one direction, your algorithm would get all thrown off. And I'm likely to do much more than just send off one scout in a straight line. The auto-explore algorithm needs to be able to take into account all sorts of funny shapes and lines in its 'way.' And on smaller maps, at least, it would get pretty hung up due to the rectangularity of the world... 

I don;t see how the search algorithm fails in retagular worlds..

If you are told to auto explore after running a long way you'll end up doing an arc one way, then turning around and doing an arc the other way, (or just jumping accross the gap),  it would need to do an edge detect, and not try to walk off the map. . . .  but that's one condition.

comparied to the alternitive, 'go to the nearest unseen patch'  i fail to see how this method could preform significantly less well.

Note that the auto explore takes the unites starting point as the expand from, not the nearest town, or center town.

can you clarify a situation where the blind go to nearest dark would be better beyond 4 or 5 turns?


Robbie Price.

Well it's more problematic than that. The spiral would essential have to start over every time the unit encounters an obstacle, and as you pointed, your algorithm only really becomes efficient in the long term... If it is constantly starting all over again, it will rarely have an opportunity to enter its efficient stage.

I'm not saying SD should stick with their simplistic and inefficient exploration algorithm, just that I'm not sure how well this would actually work in practice. That said, it could probably be used as a good starting point.

I can't discuss algorithms, but I can discuss from a results-oriented perspective.

What I'd like to see is a scout that spirals out in increasing concentric loops around it's starting point.  The 'starting point' being the nearest unexplored or otherwise visible map node.

I need to know the immediate surrounds before I want to see a scout bee-line outward to an extent then run a grid or criss-cross for a bit, then bee-line in on a tangent or right-angle before beginning another criss-cross hatch.

The CompSci majors amongst us would vomit at the inefficiency when looking at time-to-complete a given world map exploration.  But this isn't a mathematic puzzle to solve.  Reconnaissance has a different set of priorities that aren't usually factored into scout unit automation in the games I've played.  I'd like to see exploration AI in Elemental take those priorities into account.

Goodmorning all,

Firstly,  I'm not acutally suggesting a fixed spiral solution,   the algorithem is simply to aim to go beyond the nearest dark spot, along the line that reveils the most dark spots.  The algorithm generates a spiral if started in on one square surrounded by darkness, but that's just one outcome.  Also the algorith only fails when the radius of the unknown wall is small,  My example was a worse case sanario, and it showed that after as little 5 turns at one step per turn, it is better.  Since being dropped in the middle of a black field is uncommon (only for the first 5 turns, or after a random teleport). I can't think of an easier algorithm that is equally sight/turn efficent.  


Secondly, Mathmatically the best way to search (the method addopted by flys. insects, and most forging animals) for sparce resourses is to go in one direction randomly, for a random distance, then search around there locally for a random period of time, then run off in a new random direction(again for a random period of time.)  This would be a differnt algorithm,  but that algorithm in part assumes that the avalibility of resourses in a place will change over time. which is less true of 4x games. 

I play Civ4 quite a lot lately, so I'm quite familiar with the challenges of exploration. Due to the lack of decent exploration AI (my scouts tend to scout the coast line, and then die to barbarians), I usualy end micromanaging my scouts.

I think that idealy there would be saveral search algorithems and the player will choose between them for each unit as he sees fit. The algorithms would be devided by their purpose (and not functionality).

• Rough exploration - Rough idea of the landscape around you, in a wide area
• Methodical exploration - Methodicaly explore the entire region, starting from a set position.
• Borders - Send the scout in a single almost linear direction. The purpose being to check the borders of your continent.
• Random - why not?

Just thought I'd bump this post as nobody commented on the second half of the OP. I think that having different options for how multiple units stacked together move is a really good idea and would potentially be very easy to implement (at least I think it would, but I'm not a programmer :D) while offering a worthwhile increase in strategic options and cutting out some micro (I mean techincally you could do this on a unit by unit basis in most games, but having it done for you on a semi-automated basis would be a lot nicer).

I'm not entirely sold on the bonuses to fast/slow units being stacked with some of the other sort.. I can see how you could justify it from a realism point of view, but I dunno.. I'm not sure what kinda incentive structure it might set up for troop movement in general. It's still intriguing enough to be something I'd like to see beta tested however as it might potentially add some interesting decisions to the game.