Aller au contenu

Photo

AI pathfinding - the early days


  • Veuillez vous connecter pour répondre
1 réponse à ce sujet

#1
Chairon de Celeste

Chairon de Celeste
  • Members
  • 720 messages
Whenever you watch for instance a npc run into a wall and repeat that action  until
judgement day, something is wrong with her / his pathfinding.

In the early 1990ies engeneers at a university (forgot if was MIT or an european
university) faced a similar problem:
Their  tiny robots, about 10 or 15 in number had the objective to fetch items,
carry them to a location and drop  them on a pile.
The poor things just bumped into each other and never compteted their task.

The more sophisticated the program patches became, the more devastating the
results were.

Until the proverbial lightbulb flashed over the lead engeneer's head:

'why not just "tell" them to do their job avoiding any collison!'

'Small is beautyful' worked for that progam like a charm:

The robots cooperated like a well organized ant hive and the pile
on wich they dumped their load almost was a piece of art in the end.

(edit: corrected a few grammar errors and typos)

Modifié par Chairon de Celeste, 14 juillet 2010 - 04:39 .


#2
Xzel

Xzel
  • Members
  • 17 messages
Haha, this can relate to early 2000 games. I remember playing Black Hawk Down with a basic AI in player modded maps. The AI did not detect collisions so they would follow you the way they were meant to follow you, the shortest path available which is straight. This caused AIs to get stuck behind buildings if you made sharp turns or didn't check to see if the AI were following you. A very slow process but people were able to script the game's AI to detect collisions.