基于多机器人竞争行为的分布式k-WTA算法

On distributed k -WTA algorithm for competitive behaviors of multi-robots

【目的】为解决通信受限的智能体网络中的任务分配问题,提出了一种基于二次规划的通用分布式k-WTA(k-winners-take-all,k-赢者通吃)算法,本算法不需要中心命令来识别k个赢家。【方法】首先,结合高通一致性滤波器在现有集中式模型的基础上构造出一种新的分布式k-WTA模型;然后,利用拉塞尔不变性原理(Lasalle's invariance principle)计算出本模型在不变集上等价于现有集中式模型,在理论上证明了其全局渐进收敛到k-WTA问题的解;最后进行仿真试验以验证其有效性。【结果】本模型具有全局渐进收敛和智能等优势。此外,通过静态输入和动态输入的两个数值算例验证了本模型在分布式网络中搜索赢家的有效性。【结论】本研究提出的k-WTA模型能有效地解决通信受限的智能体网络中的竞争问题,可以为多机器人分布式任务分配工程应用提供参考。

[Objective]To solve the problem of task allocation in the agent network with limited communication,a general distributed k-winners-take-all(k-WTA)algorithm was proposed on the basis of quadratic programming,for which no center command was needed to identify the k winners.[Method]First,a new distributed k-WTA model was constructed based on the existing centralized model by combining the high-pass consistency filter;second,using Lasalle s invariance principle,it was calculated that the model was equivalent to the existing centralized model in the invariant set,and the global asymptotic convergence to the solution of k-WTA problem was proved theoretically;finally,simulation experiments were carried out to verify its effectiveness.[Result]The model has the advantages of global asymptotic convergence and intelligence.In addition,two numerical examples of static input and dynamic input demonstrate the effectiveness of the model in searching winners in distributed networks.[Conclusion]The k-WTA model proposed in this study can effectively solve the competition problem in the agent network with limited communication,which can provide a reference for engineering application of distributed task allocation for multi-robots.