| Session Name: | Forced-Based Anticipatory Collision Avoidance in Crowd Simulations |
| Speaker(s): | Stephen Guy, Ioannis Karamouzas |
| Company Name(s): | University of Minnesota, University of Minnesota |
| Track / Format: | AI Summit |
| Overview: | As game worlds get bigger and more populated, animating crowds of interacting characters is becoming an increasingly common task in modern computer games. Traditionally, this problem involves both rendering/animating individuals and planning their paths through the dynamic environment. Our focus is on the latter problem of computing 2D trajectories for agents, allowing them to steer towards their goals, while simultaneously avoiding upcoming collisions in an intelligent and realistic-looking fashion. The lecture will first give a brief overview of state-of the-art, velocity-based approaches (e.g., RVO and ORCA) summarizing both their advantages and drawbacks. We will then introduce a new, force-based method for anticipatory collision avoidance that follows directly from our current research in statistically analyzing human trajectories. We will show how this method avoids many of the potential pitfalls of velocity-based approaches, is easy to implement, and can be directly incorporated into existing force-based simulation pipelines. |
Recommended
