AirCapRL: Autonomous Aerial Human Motion Capture using Deep\n Reinforcement Learning

Abstract

In this letter, we introduce a deep reinforcement learning (RL) based\nmulti-robot formation controller for the task of autonomous aerial human motion\ncapture (MoCap). We focus on vision-based MoCap, where the objective is to\nestimate the trajectory of body pose and shape of a single moving person using\nmultiple micro aerial vehicles. State-of-the-art solutions to this problem are\nbased on classical control methods, which depend on hand-crafted system and\nobservation models. Such models are difficult to derive and generalize across\ndifferent systems. Moreover, the non-linearity and non-convexities of these\nmodels lead to sub-optimal controls. In our work, we formulate this problem as\na sequential decision making task to achieve the vision-based motion capture\nobjectives, and solve it using a deep neural network-based RL method. We\nleverage proximal policy optimization (PPO) to train a stochastic decentralized\ncontrol policy for formation control. The neural network is trained in a\nparallelized setup in synthetic environments. We performed extensive simulation\nexperiments to validate our approach. Finally, real-robot experiments\ndemonstrate that our policies generalize to real world conditions. Video Link:\nhttps://bit.ly/38SJfjo Supplementary: https://bit.ly/3evfo1O\n