i get essentially the same error...

The GameDev: Arena Contest
1.
a)5, lots of fun!
b)3, i put into practice a bunch of stuff that i already knew. so I definitely learned something.
c)4
d)5

2. hmm, can't really think of anything. definitely need to clear up exactly what happens when a bot uses up it's allotted time for an update...other than that i dunno. Perhaps clarify exactly what is expected in the midterm entry vs. final entry (i.e., T2K seems to think that the midterm bot should be very similar to the final entry bot. should it be? if there is to be a rule as to similarity between midterm and final entry how are you going to quantify the differences between entries?).

3. definitely not just double elim. need to have at *least* best of 5 for each bracket. the more the better...
maybe not even brackets at all...more of a pool like T2K suggests.

4. almost.

5. not that i can think of...

The GameDev: Arena Framework
1.
a)easy to do a reactive bot (5), harder to go beyond that because of certain limitations of the arena (discussed below) (2)
b)4
c)5
d)5
e)5

2.This is from a previous post of mine:
Previously, I said that i was fine with the proximity LOS not giving a definite direction, however, after playing with it a bit, I believe that it would be much better (both in terms of someone learning how to use the interface, and in terms of improving bot functionality) to stick with a perceptual scheme that is fully consistent (i.e. the proximity LOS and regular LOS should return the same types of information...should detect the same types of objects and give accurate direction/distance information). The key word here, i think, is consistancy. Either this, or if the LOS zones are going to be treated separately (in terms of the type of info returned) there should be two separate queries (GetObjectsInSight() and GetObjectsInProximity()). This would help alert people to the fact that these two queries do, in fact, return different types of info.

4.fix the problems with the arena before adding gameplay.

5.see #4

6.see #4

7.This is from a previous post of mine:
Although i think the current wall detection scheme is definitely "workable", i can see that it will probably be a major sticking point in the actual competition. The process of wall detection may be confusing to someone just trying to get into the competition. I think the ideal wall detection situation would be one in which visible wall segments are returned (i know, i know, the current query isn't set up to return this type of data...i'm just throwing out ideas). So, for example, if you were looking towards a corner and saw two walls, you would have returned to you two different wall segments. each segment would have the distance and angle to the first visible point of the wall and the dist/angle to the last visible point. As i said, this is the ideal...i have no idea how practical it is. However, as far as implementation is concerned, this could be done with a simple frustum/segment intersection test between a trapezoidal approximation of the bot's FOV and the walls.

Oh yeah, also, for everyone interested in machine learning techniques (neural nets, genetic algs, reinforcement learning, etc.) it is ESSENTIAL for there to be another mode for running the arena. Currently there is a release and debug mode. I'm gonna call this hypothetical mode learn mode. In learn mode, the arena could be run in a "fast forward" mode, and you would have access (only in learn mode!!) to information that you would not have otherwise, stuff such as (but not limited to): the enemy's health, your bot's accuracy, if your bot has collided with something, etc.

This depends on Khawks standpoint on the "fairness" of machine learning techniques in a competition like this. However, I'd like to point out that although many people tend to view these techniques as somehow magical, able to produce "ultimate" AI agents, machine learning techniques are simply that...alternate techniques. It can be just as difficult to create and tune a machine learning structure that successfully performs a task as it might be to "hard-code" a successful agent. I believe it takes just as much hard work and creativity as traditional AI techniques, so it shouldn't be disqualified on the basis that it is somehow taking the easy way out.

One major problem i have with the arena (which Tailslide, i believe, brought up) is that if part of an obstacle intersects the bots view angle (but not the center of the obstacle) your bot DOESN'T SEE IT. This results in some fairly annoying problems with obstacle avoidance/navigation. PLEASE PLEASE PLEASE fix this. one simple work around would be to include direction information on proximity objects. however, i would prefer the more elegent & consistent solution of just returning the object as seen if any part of it intersects the bot's view frustum.

there is also definitely a bug with gernades. if a grenade is in the air while another is exploding, the gernade in the air is instantly teleported to the location of the exploding gernade and explodes at the same time (this is why you are seeing bots sometimes take much more damage from a nade than expected). oh yeah, this only works for gernades fired from the same bot (i.e. if bot A's nade is about to explode and bot B fires one, this won't happen)


-------------------------------------------------------
A headache, ancillary; an hourglass auxiliary.

"something witty, blah, blah, blah"
--That One Witty Guy (remember? that one dude?)

(author of DustBot)



[edited by - drreagan on November 17, 2003 7:12:58 PM]

The GameDev: Arena Contest
1a. Yep - 5
b. Well, yeah - 3
c. 5
d. 4

2. Nothing

3. Mrm... maybe it should be done in such a way that the contestants should be grouped, and then matches are run for each group, and the top two in the group will advance to the next round. More like soccer, the world cup.

4. Yes

5. No


The GameDev: Arena Framework
1a. 5
b. 2
c. 4
d. 4
e. 4

2. Nothing

3. Nothing

4. Larger map/field of play.

5. More weapons, team battle, 8 to 16 player deathmatch on larger field.

6. Yeah. More challenging? Er.. nothing really.

7. No.

"Never met an American that didn''t like guns."
"Santa Claus would pick up a gun to save his friend."

InitGames Software | InitGames Software forums