A new method for path prediction in network games

In almost all multiplayer network games, dead-reckoning (DR) is used to predict the movements of game players, who can then predict the future movements of other players via the DR vectors they received. DR vectors, referred to as network packages, generally contain the position and velocity of game roles controlled by a sender at sending time. To achieve more accurate prediction, some games include the timestamp and acceleration of game roles in DR vectors. However, DR does not work well under bad network conditions. In our previous work [Li and Chen 2006] we proposed a solution called the interest scheme (IS), which proved to be efficient when network latency was unsteady and package loss frequent. Thus, in order to achieve much more accurate prediction, we proposed a hybrid solution. IS assumes that the path prediction for a given player is related to nearby objects or players. That is, that the players' surroundings can affect their movements, and different players may behave differently under the same conditions. In IS, a given player's surroundings are taken into account, and in order to achieve more accuracy, his habitual preferences are also taken into consideration. Experience with network games indicates that the same player will almost always behave in the same way under the same circumstances---for example, use the same fighting style. Moreover, we consider that different prediction methods should be used for different network latencies. So we introduce a hybrid method, which is a combination of IS, DR, and personal preferences. We use a 2D tank game to experiment, and compare the results of our solution with those of traditional methods. To obtain information on the players' habitual movements, we observed each participant for 30 minutes of play. Simulation shows that our method achieves significant improvements in path prediction.