Groo%20:%20Behavior-based%20Control%20of%20Real-time%20Game%20Agents

1
Groo Behavior-based Control of Real-time Game
Agents
Developers Aaron Khoo, Greg Dunham, Nick
Trienenshttp//flexbot.cs.northwestern.edu
Movement behaviors control navigation. They
affect 3 actuators rotate, translate and strafe.
We decided to control these three actuators
together since they are very closely related.

• Behavior-based AI in Half-Life Multiplayer Bots
• NPCs controlled via behavior-based techniques.
  Programs compile to parallel networks of
  finite state machines
• All inferences completely recomputed on each
  decision cycle
• Most actuators separately controlled by
  individual behaviors which are run
  simultaneously. Outputs of behaviors sent to
  FlexBot DLL on each program cycle.
• Waypoint-less navigation.
• Performance (450 Mhz Pentium3, 384 Mb RAM, 16Mb
  Riva TNT)
• 0.3 of CPU consumed per second per bot
• 52 bytes of static data memory during runtime
  per bot
• Compiles to less than 18K of object code in
  debug mode with no optimization
• Long mean time to failure. Groo bots have been
  successfully run for days at a time

Sample code for a behavior in Groo
shoot? (and facing-enemy? not-alt-fire?
(or clip-not-empty? (
current-weapon crowbar))
(or (and enemy-long-range?
long-range-weapon?)
enemy-short-range? being-shot?))
Northwestern University