Predicting Enemy Locations in Counter-Strike: Source

My research focuses on predicting opponent locations in first-person shooter video games. For my Master's degree, I developed "predictors", stand-alone modules that can be placed into a bot to provide intelligent predictive capabilities. What makes these predictors special is that they have been trained on expert human gameplay, and thus they account for common strategies and tactics of human players when prediciting enemy positions.

The video shows two predictors predicting the locations of Terrorist (T) players during a Counter-Strike: Source game on the map de_dust2. The green arrows denote Counter-Terrorist (CT) players and the teal areas are places on the map that CT players can see. On the left map, a hidden semi-Markov model illustrates its predictons by blue hue; a greater colour intesity corresponds to a higher probability of an opponent being located there. On the right map, a particle filter uses thousands of dots to represent possible places where opponents could be. The more dots in a given area, the more likely an opponent would be spotted there.

Category:

Gaming

Tags:

• Counter-Strike: Source
• Prediction
• Particle Filter
• Hidden Semi-Markov Model
• Research