Jump to content
The Dark Mod Forums

AI reactions and reasonable doubt


zergrush

Recommended Posts

So I found a bit of space to gather some numbers.

 

Here's a pic of what I measured. The granularity of the snapshots is 0.1s, not every frame, so these are rough times:

 

attachicon.gifalerts.jpg

 

This shows the amount of seconds it takes from the guard first spotting you, for him to reach each of the 5 alert levels.

 

Comparing each visibility setting in both 2.00 (back when I wrote this stuff) and 2.07 (today).

 

What pops out for me is that the delay between reaching level 4 and level 5 has dropped considerably. This delay was intended to give the player some time to duck back into the shadows.

 

What I'll do when I get more time is examine the delay code and try to figure out where the bug crept in.

 

This is very interesting! So there is a bug after all! Does it mean that in 2.0 upon reaching lvl 4 the AI would walk but not charge against you? I'm wondering if the distance compensation is also bugged at this moment, since soldiers seem to be reacting very fast from far away as well.

 

I still think, however, that the reaction times between levels 1-4, i.e. the total time between spotting the player and drawing the weapon, is a bit too fast. I think it would make more sense to increase the interval between levels 3-4 and reduce the interval between levels 4 and 5. This would mean the guard would hesitate longer before drawing his sword but would be more aggressive once it's out. It would make for a more natural reaction and would be better for stealth scores too, I feel.

Edited by zergrush
Link to comment
Share on other sites

So alert level 5 just means sword ready and in front? That's okay then! You can't even blackjack enemies with my patch then as the sword blocks it...

 

Yes, alert 5 means the AI can currently see you, which means they are attacking or attempting to close on you.

Link to comment
Share on other sites

I found the "bug", which was introduced (by me) in 2.04. The change fixed a problem first noticed in 2.04 development where a distant AI would be frozen in place if the player was too far away to "recognize as an enemy". (Think of whether you'd recognize somebody as a thief at a far distance.) The change allowed the AI to at least search around for whatever was tickling his alert level.

 

I suspect that search changes I did between 2.00 and 2.07 would also have solved the "frozen" problem, so I'm experimenting with removing the distance check.

 

Please comment on what behavior you'd expect from an AI who's far away and has spotted you in full light, raising his alert level, but not to the point where he's sure you're an enemy. Would you rely on his search pattern to bring him closer, or would you want him to delay a bit, then come straight for you?

Link to comment
Share on other sites

I suspect that search changes I did between 2.00 and 2.07 would also have solved the "frozen" problem, so I'm experimenting with removing the distance check.

 

Please comment on what behavior you'd expect from an AI who's far away and has spotted you in full light, raising his alert level, but not to the point where he's sure you're an enemy. Would you rely on his search pattern to bring him closer, or would you want him to delay a bit, then come straight for you?

 

I think the behaviour that makes the most sense would be for the AI to stare at the player, turning his head to track but not stopping if he is patrolling out of view. If the player remains in the AI's POV, the AI's alert counter will tick up until he starts walking towards the player, and then eventually running once he realizes you are an enemy (which would be when he hits alert level 3, I image)

 

Depending on how complicated we want to make it, an AI who is already alerted to an intruder should perhaps act to keep the player in his POV, which would mean stopping if he was otherwise going to walk out of view.

  • Like 3
Link to comment
Share on other sites

I found the "bug", which was introduced (by me) in 2.04. The change fixed a problem first noticed in 2.04 development where a distant AI would be frozen in place if the player was too far away to "recognize as an enemy". (Think of whether you'd recognize somebody as a thief at a far distance.) The change allowed the AI to at least search around for whatever was tickling his alert level.

 

I suspect that search changes I did between 2.00 and 2.07 would also have solved the "frozen" problem, so I'm experimenting with removing the distance check.

 

Please comment on what behavior you'd expect from an AI who's far away and has spotted you in full light, raising his alert level, but not to the point where he's sure you're an enemy. Would you rely on his search pattern to bring him closer, or would you want him to delay a bit, then come straight for you?

 

I think it would be nice to see some videos of at least a couple of proposed behavior changes in action before making any final decisions on how to permanently change the AI. Springheel's suggestion seems pretty ok though.

 

Also, perhaps we should construct a small map to test AI in the future? Some kind of map that would spawn a guard at a different distances and with different levels of lighting, perhaps.

Edited by zergrush
Link to comment
Share on other sites

Okay, I'm not getting enough responses to my questions, so let me rephrase:

 

For the given situation in Cleaning Up the Neighborhood (fully lit, far away), how long (if at all) do you think it should take for the AI to start walking toward you before he recognizes you as an enemy? In both 2.00 and 2.07, the walking starts at 2.73s for Nearly Blind.

 

Nearly Blind

Forgiving

Challenging

Hardcore

  • Like 1
Link to comment
Share on other sites

 

I think the behaviour that makes the most sense would be for the AI to stare at the player, turning his head to track but not stopping if he is patrolling out of view. If the player remains in the AI's POV, the AI's alert counter will tick up until he starts walking towards the player, and then eventually running once he realizes you are an enemy (which would be when he hits alert level 3, I image)

 

Depending on how complicated we want to make it, an AI who is already alerted to an intruder should perhaps act to keep the player in his POV, which would mean stopping if he was otherwise going to walk out of view.

 

The player is seen as an enemy at level 5, but only if he's closer than 787 units (20m). (In CutN, the "too alert" guard and the player are ~980u apart.)

 

It might be easy enough to have the AI look toward the player during level 2 (suspicious), but when he reaches level 3, he's going to start to search, and he's going to need to look where he's going.

 

As for the "complicated" suggestion: AI don't know where they are until they get there, so they won't be able to gauge "exited POV" until they've done it, which would require that they turn around and retrace their steps to regain POV. I'm not sure I want to wrap my head around the complexity of that code, since pathfinding is going to pull them each frame back onto their patrol route. I have an idea how it could be done, but I don't have the time atm to implement that much new code.

  • Like 1
Link to comment
Share on other sites

Okay, I'm not getting enough responses to my questions, so let me rephrase:

 

For the given situation in Cleaning Up the Neighborhood (fully lit, far away), how long (if at all) do you think it should take for the AI to start walking toward you before he recognizes you as an enemy? In both 2.00 and 2.07, the walking starts at 2.73s for Nearly Blind.

 

Nearly Blind

Forgiving

Challenging

Hardcore

 

I would suggest:

 

Nearly Blind: 2.5s

Forgiving: 2s

Challenging: 1.5s

Hardcore: 1s

 

One change I would make though, I think the AI should walk towards the player before drawing its respective weapon.

Link to comment
Share on other sites

To grayman's question:

In the past two years or so I've been playing, I've never been dissatisfied with the alertness of the AI. All though even on the reduced sensitivity setting they are definitely much more perceptive than the AI in other stealth games, I have never felt it represented an insurmountable challenge. The current status quo of "difficult but fair" is quite an achievement and should not be discarded lightly.

 

However, since the effect of the sensitivity sliders is pretty subtle, and because there is merit in customization, maybe it could be beneficial to adjust things a bit. Perhaps fatten up the tails as it were... Consider for example: Nearly Blind - 5s, Forgiving - 2.5s, Challenging - 1.5s, and Hardcore: - 0.5s or less. That way Nearly Blind would be comparable to other stealth games, while Hardcore would be a real challenge requiring flawless observation and planning. Of course the danger of such an adjustment is that it could radically change the way certain missions play out, depending on the player's settings. With only half a second or less of grace, I suspect large areas with lots of guards and sparse shadows might become impossible traverse without being detected. That would be unfair to people who take pleasure in ghosting missions on high difficulty, and the map makers who want to accommodate that play style.

 

On whether vision and hearing sliders should be merged:

I definitely agree they should remain separate. When I first started playing and preferred a blackjack heavy style, I cranked the hearing acuity way down while leaving vision sensitivity on high. That way I could reliably stalk down guards for blackjacking without having to worry about moss arrows or ambush point, while still enjoying the challenge of reading patrol routes to avoid line of sight. It helped me avoid frustration and continue enjoying the game until I mastered more sophisticated strategies.

 

There's also an argument about user accessibility to consider. One's ability to employ counter play against visual and auditory detection is intrinsically limited by ones own perceptiveness with these senses. Someone with a hearing impairment (or crappy headphones/speakers) might not realize a guard is nearby and can hear them until it is too late. Likewise someone with a vision impairment (or a poorly configured monitor) might not be able to notice a far off guard who can see them across a large space. Being able to tone down these mechanics individually allows such players to continue enjoying the game without sacrificing a appropriate level of challenge in the other dimensions of play they can deal with.

Link to comment
Share on other sites

Good points. Its also hard for me to judge because Ive never felt that TDM's AI are that hard, certainly not unfair. Ive noticed little difference between the previous versions and the recent one, and never experienced a situation where I thought I was targeted too quickly, I suspect these are fringe situations we are talking about. Again, Im no expert player, only played through thief 1 once and only play a bit of tdm every few months, so... Still I do agree that the full alert should be preceded by the guard noticing you, stopping and going in your direction to investigate. Since the player is fully lit, it shouldnt take much even for a far away guard to know its in fact an intruder dressed as a ninja and carrying a bow and sword.

Link to comment
Share on other sites

IMO from practical standpoint, the minimal reaction time on highest difficulty should be around 1.5s. The model presented by NH seems pretty good. Times below one sec will be impossible to tackle by 99% of the players. You'd have to have some sort of subliminal reaction mechanism to evade guards. If you don't believe me, try to measure your reaction time on something as simple as green light at pedestrian crossing. You'll be surprised how slow it is ;)

  • Like 1
Link to comment
Share on other sites

There's also an argument about user accessibility to consider. One's ability to employ counter play against visual and auditory detection is intrinsically limited by ones own perceptiveness with these senses. Someone with a hearing impairment (or crappy headphones/speakers) might not realize a guard is nearby and can hear them until it is too late. Likewise someone with a vision impairment (or a poorly configured monitor) might not be able to notice a far off guard who can see them across a large space.

That is a good argument that I didn't consider. For the responses I agree with you that it should be more than the 0.5 s difference others are recommending here. I would suggest simply:

 

Nearly Blind: 4s

Forgiving: 3s

Challenging: 2s

Hardcore: 1s

Edited by wesp5
  • Like 1
Link to comment
Share on other sites

However, since the effect of the sensitivity sliders is pretty subtle, and because there is merit in customization, maybe it could be beneficial to adjust things a bit. Perhaps fatten up the tails as it were... Consider for example: Nearly Blind - 5s, Forgiving - 2.5s, Challenging - 1.5s, and Hardcore: - 0.5s or less. That way Nearly Blind would be comparable to other stealth games

 

This is exactly what I mean by the game being far too biased towards high difficulty. Settings named "Nearly blind" or "Easy" should do what they say, and actually be nearly blind or easy, not "comparable to other stealth games" which are designed to present an appropriate challenge to a wide variety of players.

 

The underlying attitude of the Dark Mod should not, in my opinion, be "All other stealth games are pathetically easy and we need to ramp up the challenge for everybody. Git gud, scrub!". It should be "We are going to provide a game which can provide a decent range of challenge for players of all skill levels".

 

For this reason, I would suggest something more exponential for the difficulty settings, e.g.

 

Hardcore: 1s

Challenging: 2s

Forgiving: 4s

Nearly blind: 8s

 

If you think that makes "Nearly blind" ridiculous, then don't play on that setting. That's why it's called "Nearly blind" rather than "Quite easy".

  • Like 2
Link to comment
Share on other sites

I support Wesp5's times. I think that if you really want Nearly Blind to last 8 seconds, then another tier should be added in between. "Relaxed" or something like it.

 

Naturally this also open the question on whether the default difficulty should remain Forgiving.

Edited by zergrush
Link to comment
Share on other sites

Or just rename the difficulties, e.g. "Forgiving" (for casual players), "Normal" (designed to be comparable with other games), then "Challenging" and "Hardcore" for the more skilled players.

 

I suspect some players are put off by the "Nearly blind/deaf" labels, assuming it a joke or a thinly-veiled insult, and that nobody is seriously expected to play on that difficulty. Certainly no player is going to assume that these difficulty settings are designed to match what commercial games would label as "Normal".

  • Like 2
Link to comment
Share on other sites

I had a couple thoughts.

 

The nearly blind/deaf options were called that because they seemed accurately descriptive. At the time (though perhaps not any more) if you set the AI to those levels you really didn't have to worry about hiding at all. Relaxed AI would not hear you unless you were on the noisiest surfaces, and you could stand right in front of them while evoking only a, "Did I see something?" response. I remember testing the settings and I could run right down a wooden hallway all the way up to a standing guard and KO him without him being alerted enough to even turn around. To me that's "Nearly deaf".

 

Could casual players find those titles insulting? I suppose they could. I'm not sure TDM has ever been a game for casual players, to be honest. But if people want to change the names, I'm fine with it. I wouldn't reuse the same names to mean different things though. If "Forgiving" is moved to the lowest difficulty, then it's confusing for people who were currently playing on "Forgiving".

 

As for the actual times, I find it hard to judge what is acceptable in the abstract. I can't imagine how much faster 3 seconds feels than 4 or 5. In general, I think the differences between the 4 difficulty levels should be significant, and not just 1 second's worth of difference between them, but without being able to test in-game it's hard to say what's best. I was pretty happy with what we had before when the system was working as intended.

  • Like 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.


  • Recent Status Updates

    • Petike the Taffer

      I've finally managed to log in to The Dark Mod Wiki. I'm back in the saddle and before the holidays start in full, I'll be adding a few new FM articles and doing other updates. Written in Stone is already done.
      · 4 replies
    • nbohr1more

      TDM 15th Anniversary Contest is now active! Please declare your participation: https://forums.thedarkmod.com/index.php?/topic/22413-the-dark-mod-15th-anniversary-contest-entry-thread/
       
      · 0 replies
    • JackFarmer

      @TheUnbeholden
      You cannot receive PMs. Could you please be so kind and check your mailbox if it is full (or maybe you switched off the function)?
      · 1 reply
    • OrbWeaver

      I like the new frob highlight but it would nice if it was less "flickery" while moving over objects (especially barred metal doors).
      · 4 replies
    • nbohr1more

      Please vote in the 15th Anniversary Contest Theme Poll
       
      · 0 replies
×
×
  • Create New...