eduzhai > Applied Sciences > Engineering >

UAV Path Planning for Wireless Data Harvesting A Deep Reinforcement Learning Approach

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: Autonomous deployment of unmanned aerial vehicles (UAVs) supportingnext-generation communication networks requires efficient trajectory planningmethods. We propose a new end-to-end reinforcement learning (RL) approach toUAV-enabled data collection from Internet of Things (IoT) devices in an urbanenvironment. An autonomous drone is tasked with gathering data from distributedsensor nodes subject to limited flying time and obstacle avoidance. Whileprevious approaches, learning and non-learning based, must perform expensiverecomputations or relearn a behavior when important scenario parameters such asthe number of sensors, sensor positions, or maximum flying time, change, wetrain a double deep Q-network (DDQN) with combined experience replay to learn aUAV control policy that generalizes over changing scenario parameters. Byexploiting a multi-layer map of the environment fed through convolutionalnetwork layers to the agent, we show that our proposed network architectureenables the agent to make movement decisions for a variety of scenarioparameters that balance the data collection goal with flight time efficiencyand safety constraints. Considerable advantages in learning efficiency fromusing a map centered on the UAV s position over a non-centered map are alsoillustrated.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...