In this paper, we address the problem of multiple quadcopter control, where the quadcopters maneuver in close proximity resulting in interference due to air-drafts. We use sparse experimental data to estimate the interference area between palm sized quadcopters and to derive physics-infused models that describe how the air-draft generated by two quadcopters (flying one above the other) affect each other. The observed significant altitude deviations due to airdraft interactions, mainly in the lower quadcopter, is adequately captured by our physics infused machine learning model. We use two strategies to mitigate these effects. First, we propose non-invasive, online and offline trajectory re-planning strategies that allow avoiding the interference zone while reducing the deviations from desired minimum snap trajectories. Second, we propose invasive strategies that re-design control algorithms by incorporating the interference model. We demonstrate how to modify the standard quadcopter PID controller, and how to formulate a model predictive control approach when considering the interference model. Both invasive and non-invasive strategies show significant reduction in tracking error and control signal energy as compared to the case where the interference area is ignored.

在本文中，我们解决了多重四轴飞行器控制的问题，在该控制中，四轴飞行器的操纵非常接近，会由于气流而产生干扰。我们使用稀疏的实验数据来估计手掌大小的四轴飞行器之间的干扰区域，并得出注入物理的模型，这些模型描述了两个四轴飞行器（一个在另一个之上飞行）产生的气流如何相互影响。我们的物理注入式机器学习模型已充分捕获了主要由低空四轴飞行器引起的由空中气流相互作用引起的明显的高度偏差。我们使用两种策略来减轻这些影响。首先，我们提出了非侵入性的在线和离线轨迹重新规划策略，该策略可以避免干扰区域，同时减小与所需最小捕捉轨迹的偏差。第二，我们提出了侵入性策略，可以通过合并干扰模型来重新设计控制算法。我们演示了如何修改标准四轴飞行器PID控制器，以及在考虑干扰模型时如何制定模型预测控制方法。与忽略干扰区域的情况相比，侵入式和非侵入式策略均显示跟踪误差和控制信号能量的显着降低。