FrikBotNex

A QuakeC bot AI for Quake 1 deathmatch, evolved from FrikBot X (v0.10.2) by Ryan "FrikaC" Smith. The focus is on human emulation -- bots that look and feel like real players, not perfect aim-machines.

What's Different

FrikBotNex builds on the original FrikBot with:

• Combat state machine -- utility-scored decisions (fight, retreat, pressure, flank) replace binary fight-or-run logic. Skill 2+ bots evaluate situational scores with hysteresis to avoid jitter.
• Opponent modeling -- skill 2+ bots track enemy weapon, aggression, strafe tendency, and preferred distance using exponential moving averages. Used for counter-strafing, weapon counter-picks, and retreat direction.
• Human emulation across three axes:
  • Aim: flick-then-settle on target acquisition, low-frequency tracking drift via sine wave, overcorrection modeling with anti-correlated jitter
  • Movement: navigation path jitter, bunny hop fumbles, skill-scaled strafe timing
  • Decisions: reaction delay before engaging new enemies, weapon commitment locks, RL->LG combo with human-plausible success rate (40%)
• Team coordination -- bots read ally entity state to focus fire, rush enemies chasing retreating teammates, and bias retreat toward allies for crossfire
• Personality system -- balanced, aggressive, defensive, and camper archetypes that influence weapon preference, strafe speed, retreat threshold, and utility scoring
• Skill scaling -- skill 0 bots panic and fumble; skill 3 bots counter-strafe, combo weapons, and seek health while retreating. Each skill level unlocks features rather than just tightening aim.
• Navmesh navigation -- a Recast/Detour navmesh is built at map load from the BSP's hull-1 collision geometry, so bots navigate any map with no pre-authored waypoints. Off-mesh links cover everything the raw mesh can't express: jumps, drops, rocket jumps, teleporters, platforms, trains, and doors. A connectivity harness validates every item and spawn is reachable on every stock map.

Building

Compile with any QuakeC compiler from src/qc/:

cd src/qc
fteqcc

Or use gmqcc progs.src or qcc. Output goes to the repo root: progs.dat (game bytecode). A pre-compiled progs.dat is checked in for convenience.

Bot navigation runs in a custom engine build (src/server/, NetQuake + Recast/Detour as git submodules) rather than a generic Quake client/server -- see src/build/build-server.sh. Rebuild it whenever src/server/*.cpp/*.h/net_bot.c change; QC-only changes don't need it.

There are no automated tests -- testing requires running bots in this engine build.

Source Layout

src/
  qc/           Base Quake game code (GPL v2, id Software)
    defs.qc       Global declarations (must compile first)
    progs.src     Compilation manifest
    ...           Standard game modules (weapons, items, monsters, etc.)

  frikbot/      Bot AI system (public domain, Ryan "FrikaC" Smith + contributors)
    bot.qc        Entity lifecycle, declarations, main loop hooks
    bot_ai.qc     Target/enemy selection, goal management, aim pipeline, per-frame think
    bot_fight.qc  Combat movement state machine, opponent modeling
    bot_goal.qc   Item goal scoring: utility-weighted pickups, resource pools, personality
    bot_move.qc   Movement command generation, bunny hopping, off-mesh link traversal
    bot_misc.qc   Utility functions

  server/       Navmesh navigation (C++, built with the engine, not the QC progs)
    nav_hull.cpp  Extracts hull-1 BSP collision geometry
    nav_mesh.cpp  Recast/Detour navmesh build
    nav_bot.cpp   Off-mesh link detection + QC-facing path/link builtins
    net_bot.c     Fake network driver -- bots run through real engine physics

  tools/        nav_harness.sh + nav_triage.py -- full-map connectivity test gate

Installation

To integrate into a QuakeC mod, add the bot files to progs.src after defs.qc, then hook into the game loop at four points:

1. BotInit() in WorldSpawn() -- spawns physics entities
2. BotFrame() in StartFrame() -- per-frame AI processing
3. BotPreFrame() in PlayerPreThink() -- early bot checks
4. BotPostFrame() in PlayerPostThink() -- late-stage updates

See the detailed instructions at the top of src/frikbot/bot.qc.

Skill Levels

Skill	Aim	Movement	Decisions
0	High error, slow tracking, 0.3-0.7s reaction delay	Sluggish strafe, frequent bhop fumbles, nav jitter	Panic retreat, weapon commitment 2.5-4.5s, no weapon switching in combat
1	Moderate error, flick overshoot, 0.15-0.4s reaction	Faster strafe, occasional fumbles	Binary fight/retreat, team retreat bias, weapon switching
2	Lower error, tracking drift, 0.05-0.15s reaction	Quick strafe with fakes, angular retreat	Full utility state machine (fight/retreat/pressure/flank), opponent model
3	Minimal error, instant reaction	Counter-strafe from opponent model, no fumbles	RL->LG combos, health-seeking retreat, counter-picks

License

Base game code: GPL v2 (id Software). See src/qc/LICENSE.txt. FrikBot code: Public domain (Ryan "FrikaC" Smith). FrikBotNex additions: Public domain.

See ATTRIBUTIONS for full credits.