Breadcrumbs: CS 6457

Overview

Release: 1998 Developer: DreamWorks Interactive Producer: Seamus Blackley Genre: First-person action/exploration with combat Commercial Success: Disappointing, but highly influential

Key Theme: Extremely ambitious project that pushed hardware and player expectations beyond what was feasible at the time

DreamWorks Interactive History

Founding Concept:

  • Entertainment “trifecta”: movies, music, and video games
  • Founded by Spielberg, Katzenberg, and Geffen
  • Goal: Attack consumer media market from all angles
  • Intended as challenge to Disney
  • Ultimately unsuccessful in original vision
  • DreamWorks brand survived, focused on animation

Seamus Blackley Background

Expertise: Physics expert, studied at MIT

Previous Work:

  • Worked at Looking Glass Studios (no longer exists)
    • Ultima Underworld: One of first true 3D first-person games, particle dynamics for arrows/slings
    • Flight Unlimited: Real-time fluid dynamics for air movement over simulated wing (considered best flight sim controls for years)
    • System Shock: Sci-fi first-person shooter
  • Xbox evangelist (helped develop and promote Microsoft’s game console)
  • Creative Artists Agency rep
  • Active on Twitter discussing old game development

Development Issues

Major Problem: Development extended too long, DreamWorks forced completion before ready

  • Many loose ends left unresolved
  • Physics didn’t achieve stability goals
  • AI had to be gutted

Interface Design (VR-like Controls)

Philosophy:

  • No heads-up display (HUD)
  • Everything contextual in environment
  • No abstract interactions (button presses, running over collectibles)
  • Had to manipulate character’s arm to interact with objects

Health System:

  • Tattoo (heart outline) on character’s chest
  • Look down to see red ink indicating damage
  • Red ink disappears as health recovers
  • Character (voiced by Minnie Driver) speaks status

Ammo System:

  • Character verbally announces ammo count
  • Example: “About three left” after firing

Weapon Aiming:

  • No reticle
  • Must use iron sights (like real weapons)
  • Align two metal pieces (one near eye, one near barrel)
  • Very challenging in video game context

“VR” Controls:

  • “Realistic” and cumbersome inverse kinematics arm manipulation
    • Mouse controls 2D hand position relative to camera
    • Elbow bends and follows automatically
    • Modifier keys for wrist bending, grabbing, throwing
    • Hold gun in front of body, no crosshair, gun has recoil
    • Very challenging and confusing for average players

Assessment:

  • Interesting idea that would work better in VR
  • Possibly suited for Nintendo Wii
  • Made controls very challenging for 1998 players

Graphics Technology

Software Renderer

No 3D Accelerator then:

  • Used software rendering (CPU-based)
  • Required creative algorithms for performance
  • Could run at higher resolutions with modern patches

Dynamic Billboarding (Trees/Objects)

Technique:

  • Off-screen objects rendered to texture buffer
  • Trees rendered as 3D models to off-screen buffer
  • Displayed as 2D billboards in scene
  • Closest tree remains 3D model (~100 polygons)
  • Distant trees are billboards

Dynamic Updates:

  • Error metric determines when to recalculate
  • Billboard recalculates when viewing angle changes too much
  • Tree “snaps” visible when recalculating
  • Trees switch from billboard to model as player approaches

Benefits:

  • Allowed complicated rolling terrain
  • Impressive visual quality for the time
  • Efficient rendering of jungle environments

AI System (Broken)

Original Vision:

  • State machine-based AI
  • Dinosaurs work together (raptor flanking attacks)
  • Dinosaurs have changing moods/desires (hunger, satisfaction)
  • Strategy: Feed T-Rex herbivore to distract from player

Problems:

  • State machines get complicated very quickly
  • Difficult to extend with new states/behaviors
  • Each addition complicates entire implementation

Final Result:

  • Had to disable all personality aspects
  • Dinosaurs attack player on sight (exactly what they didn’t want)
  • Exception: Some herbivores leave player alone unless bothered
  • Dinosaurs programmed to never enter buildings (AI too buggy)

Level Editor Problems (3D Studio Max)

Initial Decision:

  • Use 3D Studio Max plugin architecture
  • Extend modeling software for level building
  • Seemed better than building custom tools

Critical Bug:

  • Metadata extensions stored game-specific data
  • Buffer overflow when extensions got too large
  • Undocumented size limits
  • Silent corruption: geometry appeared/disappeared randomly
    • due to extensions being too large, which caused buffer to overflow
  • Lost significant work

Solution (Kludge):

  • Empirically determined maximum metadata size
  • Daisy-chained child objects for additional storage
  • Each child provided more metadata space
  • Worked but wasn’t best solution

Lesson: Should have built their own tools instead

Physics Engine

Implementation

Bounding Box System:

  • Everything enclosed/defined by boxes
  • Compound objects: multiple boxes joined together
  • Good strategy: highly optimized collision detection
  • Focused on single primitive type

Penalty Force Method:

  • Only option given 1998 processor capabilities
  • Corrective force applied to deepest penetration point only
  • No skin offset
  • No multiple contact points
  • No iterative solving for simultaneous correction

Positive Features

Inverse Kinematics:

  • Coupled to physics system
  • Dinosaurs have ragdoll body at all times
  • Can bump into player and objects
  • Tails knock down obstacles

Ragdoll Physics:

  • First game to implement ragdoll (pioneering)
  • Dead dinosaurs enter ragdoll state
  • Can roll down hills realistically
  • Some spring system wobbling visible

Limitations

Air Hockey Effect:

  • Objects oscillate around center of mass
  • Rarely touch ground (vibrate constantly)
  • Poor friction modeling
  • Objects slide unrealistically
  • Cannot stack crates (slip off each other)
  • Dinosaurs moved weirdly

Object Interpenetration:

  • Fast-moving objects can get stuck
  • Different parts want to be on opposite sides of wall
  • Fighting back and forth between sides
  • Example: Gun stuck in wall, never recovered

Impact on Gameplay:

  • Originally intended constructive puzzles (stacking, bridges, staircases)
  • Had to cut nearly every puzzle requiring stacking
  • Physics only reliable for destructive gameplay (knocking things over)
  • Level designers had already built levels assuming stable physics
  • Extremely frustrating for players who tried to stack

Sound Design (Real-time Foley Effects)

Concept: Digital version of traditional Foley artists

Traditional Foley:

  • Sound effects artists create sounds with props
    • Coconut shells = horse hooves on cobblestones
    • Dropping glass = broken glass sounds
    • Metal sheets = thunder
  • Originated in live radio dramas

Trespasser Implementation:

Database System:

  • Collision callbacks trigger sound lookup
  • Material properties determine sound (metal on stone, etc.)
  • Multiple versions of each sound type (3-4 variations)
  • Random selection for variety

Dynamic Characteristics:

  • Energy mapping: mass × speed = volume
  • Higher energy = louder sounds
  • Lower energy = softer sounds
  • Pitch shifting for additional variety

One of coolest contributions, impressive audio system

Graphics: 3D Acceleration Problem (Biggest Issue)

Situation:

  • High system requirements at launch
  • Gamers just got 3DFX Voodoo graphics cards
  • Voodoo provided amazing performance for Quake
  • Customers expected all games to support 3DFX

Voodoo Limitations:

  • Very primitive capabilities
  • Only bilinear texture filtering
  • Limited configuration
  • No off-screen rendering
  • No custom buffers
  • No custom shader passes
  • Fixed lighting only

Trespasser’s Advanced Needs:

  • Bump mapping
  • Image caching for dynamic billboards
  • Extra lighting layers
  • Shadow mapping
  • All impossible on Voodoo hardware

Outcome:

  • Team wrote Voodoo renderer anyway
  • Killed many cool features
  • Most customers only tried Voodoo version
  • Software renderer actually looked and played better
  • Game designed for software renderer
  • This issue, on top of everything else, doomed the game

Virtual Texturing (Advanced Feature)

Concept: Virtual memory applied to texture management

How It Works:

  • Similar to OS virtual memory for processes
  • Geometry refers to virtual texture memory
  • Textures swapped based on what’s visible
  • Don’t need castle textures when in dungeon
  • Don’t need dungeon textures when at castle

Mipmaps:

  • Multiple resolution levels of same texture
  • Lower resolution for distant objects
  • Reduces aliasing (ants marching effect)
  • Can use lower resolution when bandwidth limited

Graceful Degradation:

  • Memory manager tries to load correct textures
  • Falls back to lower mipmap levels if needed
  • Catches up when memory bandwidth available

Similarity to Megatexturing:

  • John Carmack used similar concept in later Quake games
  • Best known in game Rage
  • Massive texture space with virtual management
  • Trespasser had conceptually similar system first

Technology Contributions Summary

Pioneering Features:

  • Inverse kinematics with physics
  • Ragdoll physics (first game)
  • Real-time Foley sound effects
  • Dynamic billboarding for trees/objects
  • Virtual texturing system
  • Ambitious physics engine for 1998

Lessons Learned:

  • Importance of playtesting early in development
  • Technology expectations must fit gameplay
  • Overly ambitious projects risk failure
  • Need to balance innovation with feasibility
  • Custom tools often better than extending third-party software

Legacy: Highly influential despite commercial failure, pioneered many technologies used in modern games