Multi-agent pursuit coalition formation based on a limited overlapping of the dynamic groups

摘要

Coalition formation algorithms based on organizational modeling frameworks can be considered as the earliest approaches applied to the Pursuit-Evasion problem. In this paper we have based on an extension of AGRMF (Agent-Group-Role-Membership-Function) organizational model with a limited overlapping degree of the pursuit groups to allow the coalition of the pursuers. The limited overlapping provides certain equilibrium between the pursuit groups through the elimination of negative externality problem. Regarding the path planning we have based on Markov Decision Process principles (MDP) resolved via the application of value iteration algorithm. To avoid the obstacles encountered, we have based on the recent Reward Bug Algorithm (RBA). This algorithm is based on the payoff returned by the application of MDP. The simulation results reflect the improvement provided by this new approach in relation to recent methods applied to this kind of problems.