航空学报 > 2019, Vol. 40 Issue (10): 323056-323056   doi: 10.7527/S1000-6893.2019.23056

具有通信约束的分布式SOR多智能体轨迹估计算法

卢虎, 蒋小强, 闵欢   

  1. 空军工程大学 信息与导航学院, 西安 710077
  • 收稿日期:2019-04-03 修回日期:2019-05-20 出版日期:2019-10-15 发布日期:2019-07-15
  • 通讯作者: 蒋小强 E-mail:18700827816@163.com
  • 基金资助:
    国家自然科学基金(61473308)

Distributed SOR multi-agent trajectory estimation method with communication constraints

LU Hu, JIANG Xiaoqiang, MIN Huan   

  1. College of Information and Navigation, Air Force Engineering University, Xi'an 710077, China
  • Received:2019-04-03 Revised:2019-05-20 Online:2019-10-15 Published:2019-07-15
  • Supported by:
    National Natural Science Foundation of China (61473308)

摘要: 针对传统多智能体轨迹估计算法信息交换量大,计算量随群规模指数增长,可扩展性差等诸多不足,提出了一种基于超松弛迭代(SOR)的分布式多智能体轨迹估计算法,通过将最大似然(ML)准则下的轨迹估计转化为两级线性优化问题,并综合利用分布式超松弛迭代(Distributed SOR)和标记初始化方法,加快求解速度并简化信息交换流程,最终实现了多智能体位姿轨迹优化和协作定位。实验表明,所提的分布式方法能达到集中式算法的精度水平,在49个智能体规模条件下,位置估计误差小于0.15 m,姿态估计误差小于0.03°,且数据交换量仅到现有主流分布式方法DDF-SAM的0.06%,能很好用于大规模集群的场景。

关键词: 多智能体, 分布式轨迹估计, 超松弛迭代法, 极大似然估计, 协同定位

Abstract: A fully distributed Maximum Likelihood (ML) trajectory estimation method based on the Successive Over-Relaxation (SOR) is proposed to estimate the 3D trajectories of multiple collaborative robots from relative pose measurements, which can minimize the amount of exchanged information and scales well to large teams. First, the trajectory estimation is approximated by a sequence of two quadratic sub-problems. Then, these two sub-problems are further re-parameterized into two linear optimization problems. Finally, the two linear problems are solved in a distributed manner, using the distributed SOR algorithm with flagged-initialization to reduce the amount of data transmission. Extensive experiment shows that the proposed method can reach the accuracy level of the centralized algorithm. The position estimation error is less than 0.15 m and the rotation error is less than 0.03° under the condition of 49 robot scales. What's more, the minimum data transmission of the proposed method is only 0.06% of DDF-SAM. Therefore, the proposed method can be well adopted to the application scenario of large-scale teams.

Key words: multi-agent systems, decentralized trajectory estimation, successive over-relaxation, maximum likelihood, cooperative localization

中图分类号: