Perceptions framework refinements

[Try]

Followup to discussing here: https://github.com/Try/OpenGothic/pull/589

Relevant videos: https://github.com/Try/OpenGothic/assets/100468436/2d1c819b-b7b0-4162-b3ff-2080be5a8623

This behavior observed by @thokkat, with .senses = 0, overridden in Perception.d.

In marvin g there is P.detect output. From testing it looks like this shows yes if a perc is triggered with player as source. Walking only one step at a time seems to trigger AssessQuietSound only every second step which would explain the above.

In OpenGothic, entering building already emits PERC_ASSESSENTERROOM and PERC_ASSESSQUIETSOUND is wired to ground-sounds, and then feed into senses-framework at WorldObjects::tick. Naturally those perc's are emitted only by player.

[Try]

Current idea is to:

• bypass senses framework, for PERC_ASSESSENTERROOM/PERC_ASSESSQUIETSOUND and instead loop overs npc's in nearby.
• remove SENSE_HEAR process from canSense

[Try]

In G1 npc also react inaccurate (or rather do not react) on same PERC_*

[Try]

Both cases should be fixed now, by 42770ae

One other case I'm not sure about is G1 script:

func void B_AssessQuietSound ()
{
    if (Npc_GetDistToItem(self, item) > PERC_DIST_DIALOG)   // what item?
    {
        return;
    };
    ....

If player takes item, item wont exist anymore. Maybe bug, maybe this is case of game-logic bleed into UI

[thokkat]

bypass senses framework, for PERC_ASSESSENTERROOM/PERC_ASSESSQUIETSOUND

With senses=0 a lot of passive percs if not all could be triggered. In Percption.d there is a comment

Passive perceptions are sense-independent (they always go to all NPCs, regardless of whether they are seen/heard or not)

I know comments are sometimes wrong/outdated but here it would support the testing results.

in opengothic npc react only if aiQueue is empty and there is no pending waits (aiWait)

This was introduced in https://github.com/Try/OpenGothic/commit/47c552a77a41fad65b25504b8fbcba36d66214bc to fix npc waking up while sneaking. What was working was using walkmode instead of bodystate. This prevents sending perc if e.g. opening a chest.

- if(ev.groundSounds>0 && isPlayer() && (bodyStateMasked()!=BodyState::BS_SNEAK))
+ if(ev.groundSounds>0 && isPlayer() && (wlkMode&WalkBit::WM_Sneak)!=WalkBit::WM_Sneak)
  world().sendPassivePerc(*this,*this,*this,PERC_ASSESSQUIETSOUND);

One other case I'm not sure about is G1 script

From testing: When just walking around item was never valid making function return early. If dropping something item refers to it.

[Try]

Passive perceptions are sense-independent (they always go to all NPCs, regardless of whether they are seen/heard or not)

Hm, G1-scripts? Can find this lines, for some reason. Do you know what passive mean in regard to perception? Afair at least PERC_ASSESSPLAYER/PERC_ASSESSENEMY cannot work across the wall, so they have to respect distance and ray-cast

[thokkat]

Hm, G1-scripts?

G2 scripts, comments after FUNC VOID InitPerceptions().

Do you know what passive mean in regard to perception?

Asked that myself too. Found explanation by Ikarus creator some time ago:

A distinction is made between active (searching) and passive (event-driven) perception. Active perception is a perception that is constantly running and checks itself whether something interesting is happening (a prominent example is PERC_ASSESSPLAYER, the constant observation of the player). You could say that the perception triggers a reaction on its own. Passive perception is a perception that is triggered by very specific events (damage, killing other npcs, mob movement, taking items, etc.). You could say that an event triggers the perception that triggers a reaction. (Strictly speaking, all perceptions can be seen as passive perceptions).

source: https://forum.worldofplayers.de/forum/threads/747056-Konstanter-Abstand-zu-einem-NPC?p=11695762&viewfull=1#post11695762

As I understand it: What's handled in Npc::perceptionProcess are the active ones with ASSESSITEM/ASSESSFIGHTER missing/unused, others are passive.

Afair at least PERC_ASSESSPLAYER/PERC_ASSESSENEMY cannot work across the wall, so they have to respect distance and ray-cast

When I tested with smell sense only at Harald location AssessPlayer function triggered for thief guild npcs and Gritta in Thorben's house. See sense only should behave how you described, has to be in viewing angle + visibility.

[Try]

Asked that myself too. Found explanation by Ikarus creator some time ago:

Found a better explanation in G2-scripts:

//
//  PERCEPTIONS ( ACTIVE )
//
CONST INT PERC_ASSESSPLAYER  = 1;
CONST INT PERC_ASSESSENEMY   = 2;
CONST INT PERC_ASSESSFIGHTER = 3;
CONST INT PERC_ASSESSBODY    = 4;
CONST INT PERC_ASSESSITEM    = 5;
...
//
//  PERCEPTIONS ( PASSIVE )
//
CONST INT PERC_ASSESSMURDER  = 6;
// all else

[thokkat]

Two things observed during testing:

Npcs wake up if player is picking a lock in sneak mode. Could be resolved by the mentioned above, use walkmode instead of bodystate for sneak testing.

AssessEnterRoom not sent while sneaking is incorrect. Test case can be house behind Zuris. Guard reacts on enter if sneaking. Openend https://github.com/Try/OpenGothic/pull/618 which fixes a bug that prevented this test case from working in OpenGothic.