Impact of social structure on teams of collaborating autonomous vehicles

Projects

Ray tracing

Project goals

The objective of this research is to determine whether social interactions and social structures can be used to improve the performance of teams of autonomous vehicles trying to achieve a common goal.

Approach

Given a team of autonomous vehicles (AVs), we study the impact of social rules on the ability of the team to accomplish a common goal. For the experiments described here, we use a scenario called the opera problem. In this scenario, the playing field is divided into two regions. A wall is placed between the two regions with a gap that is small enough that only a few AVs can pass through at any time. The AVs begin on one side of the wall and their goal is to move to a target location on the other side. All of the AVs are assigned the same target point. In addition to reaching the target point, it is also desirable to minimize the number of collisions. Thus, the AVs must organize themselves so that they can pass through the gap as quickly as possible without hitting each other or the walls. The performance of the team of AVs was evaluated with and without organizing social rules. Also, the difficulty of the problem can be varied by varying the size of the gap between the two regions of the simulation field as well the shape of the passage between them; a long twisting corridor is more difficult to navigate than a straight hallway. We model social interactions in terms of a dominance structure. Each AV is assigned a social caste. Individuals defer to members of equal or higher castes and ignore members of lower castes. We test four different dominance structures:

No social rules All AVs have equal dominance. (One caste.)
One dominant AV One AV has higher dominance than all other AVs. All other AVs have equal dominance. (Two castes.)
Two dominant AVs Two equally dominant AVs have higher dominance than all other AVs. The remaining AVs have equal dominance. (Two castes.)
Uniquely dominant AVs All AVs have a unique dominance value. (Number of castes is equal to the number of AVs.)

Our simulator uses the MASON simulation library developed at George Mason University.

Preliminary experiments

The following parameter settings are used in the experiments described here.

Simulation Environment Size: 300x300
Target Location: (150, 250)
AV Radius: 4
Sensor Radius: 30
Number of Directions: 72
Number of AVs: 30
Environment: Long Hall (width 60, 85), Zigzag Hall (width 75, 100)

Example runs

Long Hall, Width = 60

Publications

Adam Campbell, Annie S. Wu, Keith Garfield, Randall Shumaker, Sean Luke, and Kenneth A. De Jong (2006). Empirical study on the effects of synthetic social structures on teams of autonomous vehicles. In the Proceedings of the IEEE International Conference on Networking, Sensing, and Control, Fort Lauderdale, FL, April 23-25, 2006. [pdf]

Keith Garfield, Annie Wu, Mehmet Onal, Britt Crawford, Adam Campbell, and Randall Shumaker (2005). The effectiveness of transferring multi-agent behaviors from a learning environment in the presence of synthetic social features. In the Proceedings of the ASME Internation Mechanical Engineering Congress and Exposition, Orlando, FL, November 5-11, 2005. [pdf]

Additional information

IEEE International Conference on Networking, Sensing, and Control 2006 presentation [ppt]

Adam Campbell's Website Impact of social structure on teams of collaborating autonomous vehicles
Home Publications CV Applets Game of Life 1D Continuous CA Cyclic CA Discrete CA Totalistic CA Replacement fractal Bouncing balls Lindenmayer Blvd Genetic art Hot dogs and Pi Projects Team social structures Evolving goal priorities NEATALS Sensor/emitter sim. Ray tracing Primitives Lights Materials NURBS Misc. features Putting it together Travel destinations Videos Links	Project goals The objective of this research is to determine whether social interactions and social structures can be used to improve the performance of teams of autonomous vehicles trying to achieve a common goal. Approach Given a team of autonomous vehicles (AVs), we study the impact of social rules on the ability of the team to accomplish a common goal. For the experiments described here, we use a scenario called the opera problem. In this scenario, the playing field is divided into two regions. A wall is placed between the two regions with a gap that is small enough that only a few AVs can pass through at any time. The AVs begin on one side of the wall and their goal is to move to a target location on the other side. All of the AVs are assigned the same target point. In addition to reaching the target point, it is also desirable to minimize the number of collisions. Thus, the AVs must organize themselves so that they can pass through the gap as quickly as possible without hitting each other or the walls. The performance of the team of AVs was evaluated with and without organizing social rules. Also, the difficulty of the problem can be varied by varying the size of the gap between the two regions of the simulation field as well the shape of the passage between them; a long twisting corridor is more difficult to navigate than a straight hallway. We model social interactions in terms of a dominance structure. Each AV is assigned a social caste. Individuals defer to members of equal or higher castes and ignore members of lower castes. We test four different dominance structures: No social rules All AVs have equal dominance. (One caste.) One dominant AV One AV has higher dominance than all other AVs. All other AVs have equal dominance. (Two castes.) Two dominant AVs Two equally dominant AVs have higher dominance than all other AVs. The remaining AVs have equal dominance. (Two castes.) Uniquely dominant AVs All AVs have a unique dominance value. (Number of castes is equal to the number of AVs.) Our simulator uses the MASON simulation library developed at George Mason University. Preliminary experiments The following parameter settings are used in the experiments described here. Simulation Environment Size: 300x300 Target Location: (150, 250) AV Radius: 4 Sensor Radius: 30 Number of Directions: 72 Number of AVs: 30 Environment: Long Hall (width 60, 85), Zigzag Hall (width 75, 100) Example runs Long Hall, Width = 60 No social rules One dominant AV Two dominant AVs Uniquely dominant AVs ZigZag Hall, Width = 75 No social rules One dominant AV Two dominant AVs Uniquely dominant AVs Long Hall, Width = 85 No social rules One dominant AV Two dominant AVs Uniquely dominant AVs ZigZag Hall, Width = 100 No social rules One dominant AV Two dominant AVs Uniquely dominant AVs Publications Adam Campbell, Annie S. Wu, Keith Garfield, Randall Shumaker, Sean Luke, and Kenneth A. De Jong (2006). Empirical study on the effects of synthetic social structures on teams of autonomous vehicles. In the Proceedings of the IEEE International Conference on Networking, Sensing, and Control, Fort Lauderdale, FL, April 23-25, 2006. [pdf] Keith Garfield, Annie Wu, Mehmet Onal, Britt Crawford, Adam Campbell, and Randall Shumaker (2005). The effectiveness of transferring multi-agent behaviors from a learning environment in the presence of synthetic social features. In the Proceedings of the ASME Internation Mechanical Engineering Congress and Exposition, Orlando, FL, November 5-11, 2005. [pdf] Additional information IEEE International Conference on Networking, Sensing, and Control 2006 presentation [ppt]