UniSim: An Autonomous Multi-Agent Simulation Method with Intelligent Perception

Multi-agent simulation has become a significant method to assist studies on multi-agent problems. While many multi-agent simulation platforms or toolkits have been extensively adopted in researches, they still lack high-level autonomy and high-dimensional observations for agent-based models, which limit their scalability and practicability. In this paper, we propose UniSim, a novel autonomous multi-agent simulation method with intelligent perception to solve the key problems in multi-agent simulation. UniSim consists of three parts: Unified Environment Modeling, Autonomous Perception, and Joint United Simulation. The Unified Environment Modeling builds a dynamic sea and sky environment in real time that unifies various agents into one space domain. The Autonomous Perception provides agents with visual perception and builds an adaptive and flexible architecture to achieve autonomous information sharing. And the Joint United Simulation unifies algorithms from different time domains and extends overall capability in complicated tasks or situations. Compared with other multi-agent simulation systems, UniSim achieves comprehensive simulation for various multiagent problems and establishes a novel and universal paradigm for researches on multi-agent gaming problems which combine visual information in decision-making.