Projects

CAROSA

(Crowds with Aleatoric, Reactive, Opportunistic, and Scheduled Actions)

Most crowd simulation research either focuses on navigating characters through an environment while avoiding collisions or on simulating very large crowds. This work focuses on creating populations that inhabit a space as opposed to passing through it. Characters exhibit behaviors that are typical for their setting. We term these populations functional crowds. A key element of this work is ensuring that the simulations are easy to create and modify. Roles and groups help specify behaviors, a parameterized representation adds the semantics of actions and objects, and four types of actions (i.e. scheduled, reactive, opportunistic, and aleatoric) ensure rich, emergent behaviors. To do this, we:

Specify the characteristics (e.g. roles, goals, constraints) of individuals or groups including their behaviors and how they might differ from other individuals.
Establish the temporal (e.g. daily) activities of such individuals or groups according to their occupations or roles.
Access a library of parameterized animated behaviors that can be selected contextually, varied statistically, applied to agents, and executed in a simulation environment.
Give the agents enough attention and perception to react to the environment and people around them.
Link the framework to Commercial- Off-The-Self (COTS) software used for scheduling, enabling non-programmers to create simulations.

Contact

For more information on the CAROSA framework, email Jan Allbeck: jallbeck at gmu.edu

Publications

Li, W., Di, Z., and Allbeck, J.M. Crowd Distribution and Location Preference. In Proceedings of CASA 2012. To Appear.
Li, W. and Allbeck, J.M. Populations with Purpose. In Proceedings of Motion in Games. Springer, pages 133-144, 2011.
Allbeck, J.M. Creating 3D Animated Human Behaviors for Virtual Worlds. Department of Computer and Information Science, University of Pennsylvania, Philadelphia, PA, 2009.
Pelechano, N., Allbeck, J. and Badler, N. Virtual Crowds: Methods, Simulation, and Control. Morgan and Claypool Publishers, San Rafael, CA, 2008.
Allbeck, J.M., Functional Crowds. In Workshop on Crowd Simulation co-located with the 23rd Annual Conference on Computer Animation and Social Agents. Saint Malo, France, 2010.
Allbeck, J.M. CAROSA: A Tool for Authoring NPCs. In Proceedings of Motion in Games. Springer, 2010, pages 182-193.
Durupinar, F., Pelechano, N., Allbeck, J., Gudukbay, U., and Badler, N. The Impact of the OCEAN Personality Model on the Perception of Crowds. IEEE Computer Graphics and Applications. vol. 31, no. 3, pp. 22-31, May/June, 2011.
Allbeck, J.M. and Badler, N.I. Simulating Human Activities for Synthetic Inputs to Sensor Systems. In Distributed Video Sensor Networks. B. Bhanu, C.V. Ravishankar, A.K. Roy-Chowdhury, H. Aghajan, and D.Terzopoulos (Eds). Springer. 2011, pages 193-206.

Movies

Overview

Our first sample scenario simulates a university environment. Here we see the characters going about their day...

Scheduled Actions

Scheduled actions are the type of actions one would put on their calendar (e.g. meetings, appointments, classes). These actions add structure to simulations and also help define character roles.

Here we show how actions for individuals and groups can be scheduled through Microsoft Outlook(c) calendars. Pushing a toolbar button saves the calendar data to a MySQL database where it is accessed and simulated by CAROSA....

Failure Recovery

Possible failures are detected and dealt with.

Here an action fails due to a lack of resources (i.e. no seat in a class), and the agent responds by performing another action...

Reactive Actions

Reactive actions add life to simulations. They promote emergent behaviors that take away the robotic feel that a simulation would have with scheduled actions alone.

Here we demonstrate that agents can react to a variety of stimuli...

Opportunistic Actions

Opportunistic actions add a structured unpredictability. These actions are based on need fulfillment.

Here we see a character working in his office until he needs to get a drink, use the bathroom, and eat...

Aleatoric Actions

Aleatoric actions are essentially stochastic actions. They are composed of sub-actions with probabilities. When an aleatoric action is executed, sub-actions are chosen according to their probabilities to fill the duration specified.

Here we see a professor giving a lecture which is composed of talking, using the computer, and writing on the whiteboard...

Roles

Character roles establish baseline behaviors and in some cases locations for the characters.

Here we see characters with a variety of roles performing their default behaviors...

Object Configurations

Action and object representations facilitate a separation between an authored scenario and a modeled environment.

In this clip, an agent has been scheduled to perform four actions in sequence. The simulation is then run on three different environments without changing anything at all in the definition of the scenario...