A three-man team from SoC led by Research Fellow Lim Yew Jin, and including undergraduates Lim Zhan Wei and Travis Ho, has clinched the top spot in the 2007 ORTS RTS Game AI Competition. Held on the sides of the scholarly Conference on Artificial Intelligence and Interactive Digital Entertainment (AIIDE), the competition is an event that brings together Artificial Intelligence (AI) researchers and students interested in real-time strategy (RTS) games. The SoC team beat 11 teams in the tournament that was held from 28 May to 1 June 2007. Other universities that had fielded teams in the competition include University of British Columbia, Canada; University of Michigan, USA; and Warsaw University, Poland.
 

Teams participating in the ORTS RTS Game AI Competition have to work with Open Real-Time Strategy (ORTS), a free real-time strategy (RTS) game engine operating in server-client mode.  Commercial RTS games typically operate in peer-to-peer mode, where the entire game state is kept on the computer of each player, and the software merely hides the invisible of the game state from the players. Such an arrangement allows players to tamper with the client software to see the entire game state and therefore gain an unfair advantage. In the server-client architecture, as adopted by ORTS, the game map is stored in the server, making map-revealing hacking difficult. This exacts a heavier toll on the AI employed in a game based on ORTS, as any advantage would have to be wrought from better engineered AI, rather than opportunistic hacking. Moreover, the open architecture of ORTS allows users to connect any client software they prefer, allowing more room for autonomous AI players to pit their strength with each other, and for the quality of their underlying engineering to be compared.  
 

The SoC team entered the competition as a logical extension of their R&D work in game AI. Specific preparations for the competition, however, did not begin till it was a mere four weeks before the start of the competition because of examinations on campus. To make up for the shortfall in time, the team decided to focus on the “Strategic Combat” and “Tactical Combat” categories in the competition. Other categories include: “Collaborative Pathfinding” and “Complete RTS Game”. “We all had experience in military training, thanks to National Service, and I have strong experience in developing adversarial reasoning programs,” Yew Jin explained. “The going was slow at the beginning as we had to start from scratch, and we only got a version that would actually play something without crashing one week later. While other teams were using the last three weeks to test their programs, we were only just beginning to code up our attempt,” he added.
 

The team also credited their win to more advanced tactical combat system, more robust pathfinding routines and better strategic combat planner.

The winning system was based on the concept of an internal simulation – the computer constructs an internal representation of the game state and assigns abstract tasks such as "Hunt for nearest enemy" or "Defend this base" to a group of units. This allowed the consideration of possibilities in the future by performing simulations. During actual execution, the simulation translated the internal representations of tasks into actual commands to be executed in the "real" game. Commenting on the advantage of the design, Yew Jin said: “Once we constructed this simulation system, it was easy to construct high-level plans such as "Search and Destroy" – this presumably gave us an edge over the normal technique of hardcoding the strategy in programs. As our abstract high-level plan is removed from the actual execution of orders, we effectively could discuss a high-level strategy over lunch, and encode it into an actual strategy for our program in less than a day.”

Yew Jin’s research interests are high-performance search, artificial intelligence in games and machine learning. He has worked extensively on adversarial reasoning in games, particularly on methods to deal with the high branching factors of adversarial reasoning for his PhD thesis. He currently focuses on R&D in developing new technologies to cope with complexities of modern computer games, such as real-time strategy (RTS) and first-person shooters (FPS).

It was Zhan Wei’s first time taking part in an RTS AI competition. However, the Computer Engineering major is no stranger to other types of IT-related competitions, having twice participated in Singapore Robotics Games in 2001 and 2002, and in National Software Competition Algorithm twice before that.

Travis has a long personal history in games development. He has been pursuing his passion of building computer games since his primary school days. Among his many game development projects is a real-time strategy game entitled “Teridian Shadow” which he co-developed with a friend. The game was awarded a place at the MILIA Game Developer Village Showcase in Cannes, France in 2002. Travis is pursuing Computational Biology studies in SoC.

Videos demonstrating the team’s AI may be found here. Information on the competition may be found here.