Improving Situational Awareness in RoboFlag

Citation

Gu, Chunhui (2006) Improving Situational Awareness in RoboFlag. Senior thesis (Minor), California Institute of Technology. doi:10.7907/9PEH-B582. https://resolver.caltech.edu/CaltechTHESIS:12122012-095010036

Abstract

Situational awareness in competitive games has started to attract increasing attention in the control community. It studies how a robot identifies, understands and predicts the significant factors around it, which is essential for effective decision making and performance in any complex and dynamic environment. In this thesis, we investigate the situational awareness problems in RoboFlag, a highly dynamic testbed that comprises a mixture of offense and defense games between two robotic teams. To improve situational awareness in RoboFlag, we want to solve two main problems. (1) Real-time position estimation given limited sensing capability. (2) Optimal decision-making strategy based on position estimation.

Monte Carlo Localization (MCL), a statistical method based on particle representations of probability densities moving sequentially in discrete time, has been shown as an effective and time-efficient method for reliable position estimation, especially when the dynamics of the system and the environment are nonlinear and non-Gaussian, such as RoboFlag. In this thesis, a dynamic weight map, Hospitability Map (H-Map), that measures the ability of a target to move and maneuver at each location of the field, has been applied to MCL to enhance the efficiency and accuracy of MCL in resampling phase. Empirical results illustrate that H-Map based MCL method improves situational awareness in Roboflag by providing reliable position prediction and enhancing decisionmaking performance.

Thesis Files