The research of swarm intelligence originated from the observation and simulation of group behaviors of simple social organisms such as ant colonies and bee colonies. Since the concept was proposed in the 1980s, it has attracted the attention of researchers in many disciplines, and has become a hot and frontier field in artificial intelligence and interdisciplinary fields such as economics, society, and biology. A group refers to a whole with a large number of active individuals. Each individual has simple and basic intelligence and can interact locally with its neighbors. Swarm intelligence is a demonstration of the collective behavior of these decentralized and self-organized systems. It originates from the local interactions of simple and limited individuals in the system. The tasks completed by a single complex individual can be completed by a group of a large number of simple individuals, and the latter tends to have more robustness, flexibility and economic advantages. Crowd intelligence uses the advantages of the group to provide new ideas for finding solutions to some complex problems without centralized control and no global mode. Intelligent unmanned system is composed of unmanned platform and several auxiliary parts. It has the ability of perception, interaction and learning, and can make autonomous reasoning and autonomous decision-making based on knowledge, so as to achieve the organic whole of the goal. The main research directions are:
1. Research on the consistency of multi-agent network systems: In the multi-agent distributed coordinated cooperative control problem, the consistency problem is the basis of cooperative coordinated control between agents, which has important practical significance and theoretical value. The so-called consistency means that a certain state of all agents in a multi-agent system tends to be consistent over time. The consensus protocol is the rule of interaction and information transmission between agents. It describes the information interaction process between each agent and its neighboring agents. When a group of agents want to cooperate to complete a task, the effectiveness of the cooperative control strategy is manifested in that the multi-agent must be able to cope with various unpredictable forms and sudden changes in the environment, and must reach a consensus on the task, which requires The intelligent body system can achieve consistency with changes in the environment.
2. Research on the controllability of multi-agent network systems: Controllability is a basic concept in modern control theory, first proposed by R.E.Kalman and others in the early 1960s. Controllability is a description of the performance that the system state can be controlled by external input. Its research provides a theoretical basis for the analysis and design of system controllers and estimators. In many cases, the system cannot be controlled as a whole. In order to meet the needs, only one part of the system needs to be controlled. The controllability of a multi-agent system refers to the control of one or several leaders in the system through external input, and in the case of interaction between individuals, the follower agent can be given arbitrarily within a limited time. The specified initial state is transferred to the specified target state. Since then, most of the research on controllability is based on this structure and concept.
3. Game theory research on multi-agent network systems: With the introduction of game theory, the direct interrelationship of multi-agents is well resolved. The equilibrium solution in the game can replace the optimal solution to obtain a relatively effective strategy. At the same time, the strategy is reasonable. The learning goals of equilibrium, coordination, and cooperation in multi-agents all involve decision-making issues between agents. The central idea of game theory is to establish a strategy interaction model for the game. In game theory, the equilibrium solution is a strategy that satisfies all game players. Combination, by showing which strategies the player will ultimately use to describe the outcome of the game, using the concepts of Nash equilibrium, Stacklberg equilibrium, and meta equilibrium in game theory to guide the agent in each iteration so that the result converges to the Nash equilibrium point, and at the same time makes every The benefit of each agent is relatively large, and the convergence speed is relatively fast.
4. Autonomous intelligent precision perception and manipulation of unmanned systems: in response to the development of autonomous intelligent platforms such as sea, land, air, and space unmanned platforms, research on collaborative sensing methods based on multi-sensor information fusion in unconstrained environments; research on large-scale scene semantics Modeling and understanding methods to achieve map construction, thorough perception and dynamic cognition of complex environments; research on the rapid and accurate segmentation, detection, positioning, tracking and recognition methods of multi-source heterogeneous sensing objects in complex scenes. Establish or use existing autonomous intelligent systems for technical verification to realize natural, precise and safe interaction and precise control in autonomous intelligent unmanned systems. UAV clusters, unmanned boat clusters, unmanned tank clusters and industrial intelligent robot clusters are typical examples at the current stage. Multiple satellites with different capabilities can also form the swarm intelligence of satellite clusters, cross-domain heterogeneous unmanned clusters Further presents a more complex unmanned system group intelligence form. Unmanned system swarm intelligence has broad application prospects not only in military fields such as coordinated reconnaissance, joint operations, and battlefield evaluation, but also in civilian fields such as regional logistics, urban security, and emergency rescue.
