arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2403.19641math.OC
收藏

作为策略式博弈的能量最优多智能体导航

Energy-Optimal Multi-Agent Navigation as a Strategic-Form Game

Logan Beaver

AI总结:

本文将多智能体能量最优导航建模为策略式博弈,通过从有限约束中选择动作并确定激活时间,将轨迹编码为实数集合,实现毫秒级最优轨迹生成。

摘要:

本扩展摘要提出了一种将多智能体系统的能量最优轨迹生成为策略式博弈的方法。利用最优控制领域的最新结果,我们证明:如果约束激活序列事先已知,则可以在毫秒级时间内生成能量最优轨迹。因此,智能体不是从函数空间中选择一个无限维动作,而是从有限数量的约束中选择其动作,并确定每个约束变为活跃的时间。此外,智能体可以将其轨迹精确编码为一组实数,而不是将其控制动作作为无限维函数进行通信。我们在仿真中展示了该算法的性能,并在一台平板电脑上用45毫秒找到了最优轨迹。

英文摘要:

This extended abstracts presents a method to generate energy-optimal trajectories for multi-agent systems as a strategic-form game. Using recent results in optimal control, we demonstrate that an energy-optimal trajectory can be generated in milliseconds if the sequence of constraint activations is known a priori. Thus, rather than selecting an infinite-dimensional action from a function space, the agents select their actions from a finite number of constraints and determine the time that each becomes active. Furthermore, the agents can exactly encode their trajectory in a set of real numbers, rather than communicating their control action as an infinite-dimensional function. We demonstrate the performance of this algorithm in simulation and find an optimal trajectory in 45 milliseconds on a tablet PC.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑