In an interview broadcast live via AiGameDev.com on Sunday, Tripwire Interactive's John Gibson talked about the challenges encountered during the development of RED ORCHESTRA 2, the sequel to the critically acclaimed Red Orchestra: Ostfront 41-45. As both the President and Lead AI Programmer on the project, Gibson's insights ranged from production and design issues all the way down to implementation details.
In particular, one key feature initially cut and then re-inserted into the game was support for tanks, which can be played both alongside infantry in the single player campaign and in tank battle mode. Since the tanks are simulated in great detail in RED ORCHESTRA 2, movement itself can be a huge challenge; "most people don't know that World War 2 tanks cannot turn in place, they can only turn when moving forward or backward," as Gibson points out.
“The tanks had to take into account their turning radius when doing their path-finding or path-following, and needed to perform maneuvers like K-turns for instance if they drove down a tight alley.”
Gibson describes a specialized system than handles the tank's overall movement, built on top the navigation system to be able to perform those K-turns. He also points out that most navigation systems are built around the assumptions of character proportions, and since tanks rarely fit into a single polygon it can cause issues for the underlying implementations.
Screenshot 1: The mechanical parts in RED ORCHESTRA 2's tanks are modeled realistically, which imposes significant challenges on the AI itself — from the pathfinding to the high-level combat behavior.
On top of the low-level details, the tank movement system also had to be able to deal with a realistic damage model.
“Some of the things we struggled with were dynamic obstacles [...] and understanding the constraints of the tank's performance, for example the speed changes going up hill at a certain angle. Tanks could also get their brakes or transmission damaged, or end up with less power so they couldn't climb steeper obstacles. The game models these mechanical details very accurately, so should the brakes get damaged on one side the tank would only be able to turn in one direction.”
In these cases, what was originally a standard path-finding problem becomes a full motion planning scenario.
In the remainder of the interview, Gibson discusses the tank's combat models, enemy selection and behavior tree that simulates each of the tanks' crew members and their correct responses in a wide variety of situations. The full replay of the interview is available to PREMIUM or PLUS Subscribers from the broadcasts page along with the slides.
Screenshot 2: Behavior Tree used by the tank's commander in RO2, available on Steam for owners of the game under the Tools section.
Gibson finishes the interview with some career advice:
“Having trolled the universe for people that are good at vehicle AI, there's a market there! Not very many people are good at it and most have avoided the challenges of solving some of these issues.”
He points out that his studio, Tripwire Interactive, is looking for applicants in Game AI. Don't hesitate to get in touch with him if you enjoyed the game or the interview!