在环境未知且复杂多变的动态场景中，网络拓扑结构动态变化，群组机器人之间动 态交互，需要通过分工合作来完成复杂的业务需求。而如何合理分配资源是群组机器人 实现合作与协作的关键。本文所有机器人均为去中心化的自治机器人，将运行所需的计 算资源，通信资源等统一抽象为资源，实际并不涉及具体资源。群组机器人状态随时间 和空间动态变化，不同类型的机器人个体相互交互，相互竞争，同时环境也随时间动态 变化。针对群体内不同类型个体之间的资源分配决策问题，本文提出动态场景下基于演 化博弈的群组机器人资源分配策略，并对不同策略信息差异下的演化影响因素进行了分 析，主要工作如下：

（1）将动态场景下群组机器人资源分配问题映射成演化博弈问题，并建立演化博 弈模型，求得演化博弈的均衡解。博弈参与方为全部的资源消费机器人，策略集为场景 内所有可选择的资源提供机器人，根据本文定义列出收益函数和效用函数，得到复制动 态方程，求得均衡解。收益函数取决于资源提供机器人的带宽和服务价格，效用函数主 要包括信息传输时间和传输能耗两部分。并针对动态场景的特性，引入记忆长度内的历 史信息，对演化博弈模型进行优化。最后对建立的模型进行数值仿真分析，结果表明在 25 次迭代后，该群体的分布趋于收敛，且稳定存在。通过对不同记忆长度下的历史平均 收益信息和当前收益信息比较，得出在中等记忆长度 m=5 时收敛速度最快，收敛性最好。 最后进行算法对比，进一步证明本文提出的基于历史信息的演化博弈模型的有效性。

（2）针对演化的影响因素，研究了资源提供机器人在不同策略信息差异下的行动 选择对于演化策略的影响。资源提供机器人在决策时，存在考虑对方行动或者不考虑对 方行动两种选择。将其分别抽象为 Stackelberg 博弈模型和古诺博弈模型。根据两种选择 策略的收益函数建立演化博弈复制动态方程，证明复制动态方程是凸函数并求得演化均 衡解。最后对演化博弈结果进行仿真测试，仿真结果表明，当两种策略的比例分别为 0.5 时，机器人会逐渐收敛到收益大的一方，此时收益大的策略为占优策略；当两种策略选 择为不同比例时，在第一轮以固定的比例进行博弈，之后加入随机性，让机器人个体在 两种策略中随机选择。结果表明，经过一段时间的突变，选择先后出价的比例最终稳定在 0.62，选择同时出价的比例最终稳定在 0.38。

本文针对动态环境下的群组机器人资源分配问题，建立演化博弈模型并进行仿真分 析，证明了所提模型的有效性，为将来的研究打下基础。

In dynamic scenarios with unknown and complex environments, dynamic changes in network topology and dynamic interactions between swarm robots require division of labor to accomplish complex business requirements. How to reasonably allocate resources is the key for swarm robots to realize cooperation and collaboration. In this paper, all robots are decentralized autonomous robots, and the computing resources and communication resources required for operation are abstracted as resources, and no specific resources are involved. The state of swarm robots changes dynamically over time and space, and different types of robots interact and compete with each other, while the environment also changes dynamically over time. To address the resource allocation decision problem between different types of individuals within a group, this paper proposes an evolutionary game-based resource allocation strategy for swarm robots in dynamic scenarios, and analyzes the evolutionary influences under the information differences of different strategies, and the main work is as follows:

(1)The resource allocation problem of swarm robots in the dynamic scene is mapped into an evolutionary game problem, and an evolutionary game model is established to find the equilibrium solution of the evolutionary game. The game participants are all the resource consuming robots, and the strategy set is all the selectable resource providing robots in the scene, listing the revenue function and utility function according to the definition of this paper, obtaining the replicated dynamic equations, and finding the equilibrium solution. The revenue function depends on the bandwidth and service price of the resource providing robots, and the utility function mainly consists of two parts: information transmission time and transmission energy consumption. And the evolutionary game model is improved by introducing the historical information within the memory length for the characteristics of the dynamicscenario.Finally, the established model is analyzed by numerical simulation, and the results show that after 25 iterations, the distribution of the group tends to converge and exists stably. By comparing the historical average return information and current return information under different memory lengths, it is concluded that the fastest speed of convergence and the best convergence is achieved at medium memory length m=5. Finally, an algorithm comparison is carried out to further prove the effectiveness of the evolutionary game model based on historical information proposed in this paper.

(2)For the influencing factors of evolution, the influence of the action choices of resource providing robots under different strategy information differences on the evolution strategy is investigated. Resource providing robots have two choices of considering each other's actions or not considering each other's actions when making decisions. They are abstracted as Stackelberg game model and Gounod game model respectively. Based on the payoff functions of the two choice strategies, we establish the replicated dynamic equation of the evolutionary game, prove that the replicated dynamic equation is a convex function and find the evolutionary equilibrium solution. Finally, the simulation test on the results of the evolutionary game is carried out, and the simulation results show that when the proportion of the two strategies is 0.5 respectively, the robot will gradually converge to the side with the larger gain, and at this time, the strategy with the larger gain is the dominant strategy; when the two strategies are chosen as different proportions, the game is played with a fixed proportion in the first round, and then stochasticity is added to let the robot individuals choose randomly among the two strategies. The results show that after a period of mutation, the proportion of choosing successive bids is finally stabilized at 0.62, and the proportion of choosing simultaneous bids is finally stabilized at 0.38.

In this paper, an evolutionary game model is developed and simulated for the resource allocation problem of swarm robots in dynamic environments, proving the effectiveness of the proposed model and laying a foundation for future research.