In this paper, we demonstrate a formulation for optimizing coupled submodular maximization problems with provable sub-optimality bounds. In robotics applications, it is quite common that optimization problems are coupled with one another and therefore cannot be solved independently. Specifically, we consider two problems coupled if the outcome of the first problem affects the solution of a second problem that operates over a longer time scale. For example, in our motivating problem of environmental monitoring, we posit that multi-robot task allocation will potentially impact environmental dynamics and thus influence the quality of future monitoring, here modeled as a multi-robot intermittent deployment problem. The general theoretical approach for solving this type of coupled problem is demonstrated through this motivating example. Specifically, we propose a method for solving coupled problems modeled by submodular set functions with matroid constraints. A greedy algorithm for solving this class of problem is presented, along with sub-optimality guarantees. Finally, practical optimality ratios are shown through Monte Carlo simulations to demonstrate that the proposed algorithm can generate near-optimal solutions with high efficiency.
翻译:在本文中,我们展示了一种优化组合子模块最大化问题的配方。 在机器人应用中, 优化优化问题是相当常见的, 优化问题是相互结合的, 因此无法独立解决。 具体地说, 如果第一个问题的结果影响到在较长时间范围内运作的第二个问题的解决办法, 我们考虑两个问题。 例如, 在我们的激励性环境监测问题中, 我们假设多机器人任务的分配将有可能影响环境动态, 从而影响未来监测的质量, 并在此以多机器人间歇性部署问题为模型。 解决这种类型的混合问题的一般理论方法通过这个激励性例子来证明。 具体地说, 我们提出一种方法来解决以亚型组合功能和甲状体制约为模型的混合问题。 提出了解决这一类问题的贪婪的算法, 以及亚型最佳性保证。 最后, 通过蒙特卡洛模拟显示实际的最佳性比率, 以显示拟议的算法可以产生近于最佳效率的解决方案 。