An Application of Model-Free Reinforcement Learning to the Control of Aerial Vehicles With Slung Payloads

Le résumé fourni par la source

Model-free control has gained in popularity in recent years for its adaptive, data-driven solutions to complex control objectives. This paper discusses the application of model-free reinforcement learning (RL) to the control of multibody aerial vehicles. First, a summary of model-free RL as it applies to robotic systems is presented. Then, a model-free RL controller is demonstrated, whereby an actor-critic algorithm is used to stabilize the swing of a slung payload carried by an autonomous quadcopter aerial vehicle, utilizing the latter’s full continuous action-space. Its effectiveness and adaptability are first demonstrated in simulation, and then validated on an experimental testbed. The algorithm is shown to be computationally efficient enough to adapt to the experimental testbed in real-time, and to work within the framework of widely-used autopilot software ArduPilot.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.