Guide to anticipatory collision avoidance

In games, the act of computing paths that reach a character’s current goal is typically accomplished using some form of global planning technique (see, e.g., [Snook 00, Stout 00]). As a character moves along the path toward its goal, it still needs to intelligently react to its local environment. While, typically, there are not computational resources available to plan paths that account for every local detail, we can quickly modify a character’s path to stay free of collisions with any local neighbors. In order to keep this motion looking realistic and intelligent, it is important that our characters show clear anticipation even for this local collision-avoidance routine. Consider, for example, the scenario shown in Figure 19.1, where two agents pass each other walking down the 19.1 Introduction 19.2 Key Concepts 19.3 Prototype Implementation 19.4 Advanced Approaches 19.5 Conclusion References same path. On the left, we see the result of a last second, “bouncy-ball” style reactionthe characters will get to their goals, but the resulting motion does not display much anticipation. In contrast, the right half of the figure shows our desired, humanlike behavior, where characters are able to anticipate the upcoming collision and efficiently adapt their motions early on.