Probabilistic Plan Synthesis for Coupled Multi-Agent Systems
It addresses the computational bottleneck of centralized controller synthesis for multi-agent systems with probabilistic specifications.
This paper proposes an automated method for synthesizing controllers for multi-agent systems under uncertainty, where each agent has individual probabilistic temporal logic specifications and may need to collaborate via shared actions. The method achieves better computational complexity than centralized approaches.
This paper presents a fully automated procedure for controller synthesis for multi-agent systems under the presence of uncertainties. We model the motion of each of the $N$ agents in the environment as a Markov Decision Process (MDP) and we assign to each agent one individual high-level formula given in Probabilistic Computational Tree Logic (PCTL). Each agent may need to collaborate with other agents in order to achieve a task. The collaboration is imposed by sharing actions between the agents. We aim to design local control policies such that each agent satisfies its individual PCTL formula. The proposed algorithm builds on clustering the agents, MDP products construction and controller policies design. We show that our approach has better computational complexity than the centralized case, which traditionally suffers from very high computational demands.