next »

[BleakProspects]

http://www.dwarfcorp.com/

[kurtss]

Give that website a gold medal, I've never seen anything as great as that. 

[Seiseki]

About the website, I'm a bit concerned that all the content might be overwhelming.
You should probably dial it down a bit..

[Eigen]

Minimal is in this fall/winter season.

[zede05]

Haha, nice name, I like it. Although I think the logo could be better, like a silhouette of the logo with the dwarves building it.

[Jorum]

The rendering reminds me of Magic Carpet (though more cube-y). Looks great!

[JigxorAndy]

This is looking really pretty! I like the new name as well. It grabbed my attention as opposed to the WIP thread name. Can't wait to see this progress more and the website take off.

[laserfire]

http://www.dwarfcorp.com/

Watch your kerning on that title.



I messed with it but it's kinda nasty not being able to work with the live type.

[Swaggermuffin]

It's gonna be a little difficult not thinking of it as "Some DF-Looking Minecrafty Thing" anymore.

[DustyDrake]

http://www.dwarfcorp.com/

Watch your kerning on that title.



I messed with it but it's kinda nasty not being able to work with the live type.

I don't see a differen- ooooohhhh.

[BleakProspects]

I'm glad that my thread is starting to look like the equivalent of a Head-On! commercial.   DWARFCORP. DWARFCORP. DWARFCORP.

In other news, this week was Performance Week I. I got around to fixing *most* of the bugs caused by switching from octrees to spatial hashing for storing chunks, and I *think* it sped up a bit, but not as much as I'd hoped. Anyway, it was necessary for scaling some things up. Building chunks is marginally faster, as it lighting, planning, raycasting, and more.

I also improved planning dramatically by doing pseudo-bi-directional A star. Now, if the plan fails, instead of just returning the best path so far (which is usually off in a bad direction), the dwarf will instead try to plan from the goal to it's current position before giving up. This fixed most of my issues with dwarves not being able to plan from outside to inside.

Now I'll focus on gameplay again. I'm trying to decide whether to redo the control scheme to a more traditional RTS scheme over the current Dwarf-Fortress-esque scheme (ie, select units and tell them what to do, rather than dwarves automatically deciding what to do based on the tasks you create). Or I might postpone this for a while and instead focus on implementing combat and death.

[BleakProspects]

There are forums now:
http://www.dwarfcorp.com/forum/

Keeping with the theme, there is nothing there... YET.

[BleakProspects]

Started implementing another creature so that dwarfs could fight it. I am quickly beginning to realize how unscalable my AI code is.

Currently, my AI is an explicit finite state machine where each state is a class I have to instantiate. A dwarf has about ~20 states. Just defining the state machine takes ~300 lines of code, and the states themselves make up over 1000 line of code.

This is going to get exponentially worse as I add more entities and states, so I think I will try something radical instead. Looking at doing Goal Oriented Action Planning, or STRIPS.

http://web.media.mit.edu/~jorkin/gdc2006_orkin_jeff_fear.pdf

No idea how to do this yet. Might be going down a dead end.

[BleakProspects]

Implemented a barebones version of the GOAP algorithm today. This may not look like much, but it's probably the most important output I've seen from the game so far!  

Goal: KillEnemy (EnemyDead = true)

Start:
HasWeapon: false
EnemyDead: false
AtArmory: false
NearEnemy: false


Plan:
AcquireTarget
PlanToArmory
AcquireWeapon
LeaveArmory
PlanToTarget
AttackTarget

GOAP (Goal Oriented Action Planning) works by giving an entity a goal world state, and a set of actions it can perform which change the state of the world. Actions have pre conditions and post conditions. It then plans in an "action-state space." This is similar to the STRIPS algorithm developed for robotics at Stanford in the 1970's. It was used for NOLP and FEAR in the early 2000's.

What's *REALLY* cool about this is that I didn't have to tell the AI what to do in every situation (which I would have to do in a state machine). The AI instead figured out *on its own* that it needed to first go to an armory, then get a weapon, then leave the armory, then attack an enemy. It actually finds the sequence of actions with the least cost that satisfy its goal. So if there were more ways to kill an enemy than by this method (say with your fists) which were more expensive, it would only pick those as a last resort.

Next I'll apply this to actual dwarf AI and start getting really intelligent behavior. Also, this has the plus of being applicable to my robotics phd research...

[poe]

Implement Maslow's Hierarchy of needs please.