BioWare Activity Pillars and Design Innovation

1
AI.Implant and BioWare Corp.Game Developers
Conference 2007 Graham Wihlidal Programmer
Sean Megaw Producer Unannounced Next-Gen
Project UE3 Tech
2

Presentation Outline

• Overview
• Why AI.Implant?
• UE3 Integration
• Workflow Productivity
• Post-Adoption Improvements
• Discussion of Various Tests
• Conclusion

3
Overview

• Working on an unannounced project using UE3.
• We have recently adopted AI.Implant for our AI
  solution.
• Still in prototype, but have already witnessed a
  number of feature and productivity gains.
• Designers are already accustomed to working in
  UnrealEd. - AI.Implant integrates smoothly into
  the toolset easy for designers to understand the
  workflow.
• Our project needed a powerful and customizable
  solution that would match the requirements for
  our next-gen world.
• We heavily utilize AI.Implants Dynamic Path
  Refinement (DPR) technology as our worlds are
  largely dynamic and dynamic path finding can be
  problematic to develop.

4
Why AI.Implant?

• Our project has very ambitious goals (especially
  AI) reduced risk
• Much more cost\time efficient to adopt a solid
  and proven middleware as opposed to developing
  our own.
• AI.Implant ships with excellent
  debugging\development tools, and the technical
  support is lightning fast.
• Very easy for designers to grasp fairly tame
  learning curve.
• Believable character behaviors. BioWare is
  renowned for story-telling and character
  development. AI.Implant is the perfect fit for
  our vision. Big active world that feels alive.

5
Why AI.Implant?

• Rapid iteration for code updates and bug fixes
  technical support is extremely helpful and
  timely.
• We set up a number of teleconferencing calls to
  work through some integration and design issues
  that came up, and the staff at AI.Implant were
  very responsive to us.

6
UE3 Integration

• Our project is using Unreal Engine 3.
• A requirement for our AI solution was that it
  had to seamlessly integrate with UnrealEd to
  provide a consistent designer-driven workflow.
• AI.Implant integrated smoothly into Unreal, and
  AI.Implant staff were always readily available to
  help answer questions.
• Code drops were provided that showed our
  developers how to integrate AI.Implant into UE3.
  Integration was fairly painless.
• We integrated into a customized version of UE3
  integrating into base UE3 is even easier. Comes
  down to replacing base Epic files with their
  version.

7
Workflow Productivity

• Seamless integration with UnrealEd designers
  feel at home with workflow (consistent with
  Epic).
• Excellent development and debugging tools. Can
  visualize AI behavior at run-time.
• Behaviors can also be exported and played in an
  external viewer to rule out the possibility that
  a bug is engine-related.
• Largely data-driven, so requires minimal
  programmer assistance.
• Plug-ins for 3ds Max Maya. Custom tool
  integration supported as well.
• Authoring manual and tutorials.

8
Workflow Productivity
9
Workflow Productivity
10
Workflow Productivity

• Adding navigation mesh nodes to a large map
  layout would normally take 6 hours.
• Using the 3ds Max exporter to automatically
  generate navigation mesh data took around 5-10
  minutes on the same data set.
• In addition to the decision tree editor, a
  language called AI.script (very similar to
  UnrealScript) is available for AI logic that is
  not easily described using a simple tree
  structure. This accessibility places even more
  power in the hands of designers.
• We hardly ever need to use script, which means
  that even if a designer cannot write scripts they
  can still remain very productive.

11
Workflow Productivity
12
Workflow Productivity

• A very beneficial feature of AI.Implant is the
  ability to change logic at run-time and view the
  changes instantly without any recompilation of
  scripts.
• Allows for rapid iteration of AI content.
• Bypasses most content pipeline bottlenecks,
  ensuring designers are always productive.
• AI behaviors are all dynamic, so you can
  visually tweak your AI until it behaves as
  expected debugging is very easy too.

13
Workflow Productivity

• AI.Implant was built with UE3 streaming in mind,
  and does an excellent job at it.
• Multiple navigation meshes can be created and
  linked together with meta-connections.
• Each navigation mesh can then be streamed in and
  out dynamically without breaking path finding.
• NPCs will path find to a meta-connection that
  links two streamable navigation meshes together,
  and then load\unload dependencies in and out as
  required.
• Huge win for performance and memory footprint.

14
Post-Adoption Improvements

• Exceptional path-finding (quality of pathing,
  performance, memory footprint, static dynamic
  environments).
• Designers new to AI.Implant can hit the ground
  running with a workflow consistent to Epic, and a
  number of useful tutorials and samples.
• Very minor programmer support needed to script
  complex behaviors. Our old system was largely
  programmer driven, which usually resulted in
  design bottlenecks.
• We found that for the map layouts in our game,
  AI.Implants DPR technology had dramatically
  better memory, performance and workflow behavior
  than base UE3 path finding.

15
Discussion of Various Tests

• We developed a factions demo that has multiple
  faction groups, and each NPC is assigned to one.
• Opposing factions will meet up, do a dice roll
  to simulate combat, and the winner will escort
  the other NPC to a jail.
• Also developed a cops and robbers demo that
  has 3 NPC classifications cop, robber, civilian.
• Civilians wander around the area.
• Robbers attempt to mug civilians and then run
  away.
• Cops chase down robbers and bring them to jail.
• Many other prototype demos created all were
  easily developed and simulated with AI.Implant.

16
Discussion of Various Tests

• AI.Implant really shines with crowd and follower
  behaviors. Performance and memory is
  exceptionally scalable, and the simulation of
  dense crowd flocking and reaction is very
  believable and responsive.
• AI behavior can be easily constrained to your
  CPU budget. You can specify how often complex
  scripts execute, and you can specify an execution
  offset for different behaviors to spread out CPU
  time.
• We are very excited about our improved and
  extended AI capabilities!

17
Conclusion

• Huge win for our project next-gen AI is
  obviously important for a next-gen title.
• Incredible savings with cost\time using
  AI.Implant instead of writing our own large scale
  AI solution.
• Bulk of AI behavior is designer driven
  unblocking designers to allow for rapid
  iteration, and allowing programmers to work on
  more low level tasks.
• AI.Implant workflow promotes reusability of
  behavior when appropriate this helps optimize
  memory footprint important for console
  development.
• We can spend more of our time purely on
  developing a great game that matches our vision,
  instead of working on the grunt work.

18
Questions? Graham Wihlidal (graham_at_bioware.com) S
ean Megaw (smegaw_at_bioware.com) Engenuity
(info_at_engenuitytech.com)