The objective of our research is to develop efficient techniques for simulating group behaviors. We investigate how agents can work cooperatively to perform tasks, plan paths in dynamic environments, or influence another group of agents to locations in an environment. Our goal is create a framework for simulating and controlling communities of characters that can dynamically interact with each other and their environment. There are many important applications of this system, ranging from civil crowd control (e.g., planning exit strategies from buildings or sporting event venues), to education and training (e.g., providing museum exhibits or training systems), to entertainment (e.g., interactive games). While there are existing methods that focus on the simulation aspect, there is a lack of methods that support the interaction and control (or steering) of multiple groups of agents.

This work focuses on a framework that addresses these challenges by integrating roadmap-based path planning with agent-based modeling. Our initial work introduced the idea of integrating roadmap-based path planning with agent-based emergent behavior. That initial work studied single group behaviors such as covering and a simple multiple group shepherding behavior and established that this hybrid approach has promise. Currently, we are extending our approach to support a greater range of scenarios that involve:

• large numbers of agents (e.g., crowd control in emergency situations),
• agents that adaptively and dynamically select among different behaviors (e.g., switching to evasive behaviors when enemies are detected), and
• dynamic and arbitrary groupings and coordination of agents (e.g., coordinated group patrols or pursuits).

Our general strategy is to integrate multi-agent simulation with roadmap-based path planning. We use a graph-based representation (roadmap) of the environment that encodes representative feasible pathways and also other important information about the environment, e.g., the locations of exits, safe areas for evacuation, clearings in forests, or hiding spots. Our framework provides a uniform way to model, select/combine, and specialize common basis behaviors to create new emergent behaviors. Improved scalability and more complex group behaviors and interactions are supported by mechanisms designed specifically to handle the modeling of dynamic group formation, intra- and inter-group interaction, and the customization of behaviors based on group membership.

Below we list some of our current focuses and provide links to pages providing more details.

Evacuation Planning
One important application of our work is evacuation planning. By being able to simulate agents that are evacuating an area, we can study the effects of things such as the number and placement of exits, how losing exits affects evacuation routes and times, or how evacuation times vary depending on the number, type and placement of barriers and directing agents available to control the evacuating agents.

In our initial work in this area, we study a scenario where some agents are attempting to evacuate the first floor of a building. The agents have to find paths to the safe areas. They use their knowledge of the environment (a roadmap) and information they learn about the situation by discovering barriers blocking routes or from directing agents (e.g., emergency response personal or posted signs) indicating which exists to use/avoid and/or which safe areas they should evacuate to.

Pursuit Evasion Techniques
Pursuit-and-evasion are commonly studied behaviors. One group of agents, the pursuers, attempts to find and capture another group of agents, the evaders. The evaders attempt to remain undetected and once detected, attempt to escape and hide from the pursuers. We look at aspects of pursuit-evasion which involve studying searching techniques, pursuit strategies and evasion heuristics.

All Behaviors

All Behaviors