Autonomous agent team : joint situation awareness and cooperative learning

Abstract

Autonomous team agents are used in real life applications such as search and rescue operation. Each autonomous agents is simulated by a software entity which is able to carry out some set of operations on behalf of its owner. This kind of agents is considered as a part of technical or natural environment in the sense that these agents are able to sense any status in the environment and accordingly in pursuit of its intended agenda. The motivation for deploying autonomous agents is to firstly increase the efficiency and secondly the success rate for conducting search and rescue missions. The objective of this research is to look into various aspects of collaboration between agents which include decision making for agent’s target and finding best targets for the mission. This project will focus on the aspect of analyzing the problem of multi-agents in collaborative problem solving and improving agent assignment to solve a scenario problem. This will involve looking into an existing solution which solves the decentralized assignment problem with the implementation of decentralized stochastic algorithm (DSA) and the project also includes implementing different variants which could affect the performance of simulation. There are several algorithms which could be used for solving the decentralized assignment problem such as the decentralized stochastic algorithm (DSA) and the distributed breakout algorithm (DBA). However this project will be focusing on the decentralized stochastic algorithm (DSA) because it was implemented in the existing solution. Specifically, an improved algorithm is made from the existing distributed stochastic algorithm solution with a newly introduced probability to decide whether an agent will change to its neighbour’s target based on a utility score value. The simulation result of the original DSA solution is served as a benchmark against the implemented solution. The results of the modified solution show a slight improvement compared to the original results in terms of the reduction in the number buildings burned. These results also indicate the improvement made for the agent changing to its neighbour’s target with the same probability defined for the decentralized stochastic algorithm (DSA).