Decentralized task re-planning approaches with en route information rewards

Abstract

This paper presents extensions to the consensus-based bundle algorithm (CBBA) to take into account information rewards obtained en route. Original CBBA gets score from a performed task when it reaches to the target area, however cannot acquire score when current defined task is changed before completion. The key idea is to incorporate acquired information on the fly when defining scores of the assigned tasks in the re-plan process so that agents can react to the changes in the environment with correct awareness of the executed task scores. Real-time operation is one of the key issues of this research, and it can be considered by linearizing the information changing using pre-defined formula. In order to adjust the algorithm for simulation in real time domain with a better performance, scoring scheme should be pre-defined as a specific function according to a path to the target. Along the way to the target, amount of information gathered differs with type of sensors as well as path of the agent. In this research, the agent’s movement is approximated as moving towards the target in a straight line with constant velocity. The acquired information is calculated using the entropy method as mutual information in the case of vehicles using the bearing sensor, range sensor or both. This pre-defined tendency of gathering information is expressed by a specific formula and reflected to the CBBA reward scheme and compared to the original CBBA. The CBBA reward scheme, which adjusts sensor characteristic, reflected more realistic situations in mission field.