Deep Learning for Robotic Mass Transport Cloaking

摘要: We consider the problem of mass transport cloaking using mobile robots. The robots move along a predefined curve that encloses safe zone and carry sources collectively counteract chemical agent released in environment. goal is to steer flux around desired region so it remains unaffected by external concentration. formulate controlling robot positions release rates as PDE-constrained optimization, where propagation modeled advection-diffusion (AD) PDE. use neural network (NN) approximate solution Particularly, we propose novel loss function for NN utilizes variational form AD-PDE allows us reformulate planning an unsupervised model-based learning problem. Our discretization-free highly parallelizable. Unlike passive methods metamaterials flux, our method first actively control concentration levels create zones independent environmental conditions. demonstrate performance simulations.