Scholarlink

<p>Unmanned Aerial Vehicles (UAVs) are quickly gaining importance in numerous applications as they can cover vast areas and reach otherwise difficult to access points. To improve overall UAV performance, this paper proposes a new conceptual framework called Adaptive Obstacle Avoidance Based UAV Trajectory Planning with Segmentation Using ConvUNext enhanced with Meta-reinforcement learning (AOTPSM). The framework includes ConvUNext, that utilizes a U-Net topology for enhanced image and video segmentation as a result of enhanced feature extraction and detailed refinement. Furthermore, this paper introduces a novel method of Dynamic Energy-Aware Path Planning based on Twin Delayed Deep Deterministic Policy Gradient (TD3) to reduce energy consumption of UAVs during long time operations by identifying the best path based on past data and environmental conditions. In the case of Adaptive Trajectory Optimization, the framework uses Proximal Policy Optimization (PPO) combined with Soft Actor-Critic (SAC) to address trajectories with a stochastic policy, a replay buffer, and twin <i>Q</i>-Value Networks to handle trajectories effectively and balance energy in challenging situations. The SAC algorithm also helps in detecting obstacles and also passes the processed data to the Model-Agnostic Meta-Learning (MAML) which enables the robot to learn quickly in a dynamic environment with little retraining. The AOTPSM framework was rigorously tested with the UDD-6 dataset, achieving a highest effectiveness of 97.0%, followed by ConvUNext at 95.2%, PPO with SAC at 94.3%, TD3 for Dynamic Energy Management at 92.5%, and SAC with MAML for Obstacle Detection at 91.8%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Energy-Aware Adaptive Obstacle Avoidance Based on Meta-Reinforcement Learning with Segmentation for UAV Trajectory Planning

  • 期刊论文

  • 发表日期：

    2025年4月23日

查看全文
  • T. Archana,
  • T. Aravind,
  • A. Hema Malini,
  • C. T. Kalaivani

摘要

Unmanned Aerial Vehicles (UAVs) are quickly gaining importance in numerous applications as they can cover vast areas and reach otherwise difficult to access points. To improve overall UAV performance, this paper proposes a new conceptual framework called Adaptive Obstacle Avoidance Based UAV Trajectory Planning with Segmentation Using ConvUNext enhanced with Meta-reinforcement learning (AOTPSM). The framework includes ConvUNext, that utilizes a U-Net topology for enhanced image and video segmentation as a result of enhanced feature extraction and detailed refinement. Furthermore, this paper introduces a novel method of Dynamic Energy-Aware Path Planning based on Twin Delayed Deep Deterministic Policy Gradient (TD3) to reduce energy consumption of UAVs during long time operations by identifying the best path based on past data and environmental conditions. In the case of Adaptive Trajectory Optimization, the framework uses Proximal Policy Optimization (PPO) combined with Soft Actor-Critic (SAC) to address trajectories with a stochastic policy, a replay buffer, and twin Q-Value Networks to handle trajectories effectively and balance energy in challenging situations. The SAC algorithm also helps in detecting obstacles and also passes the processed data to the Model-Agnostic Meta-Learning (MAML) which enables the robot to learn quickly in a dynamic environment with little retraining. The AOTPSM framework was rigorously tested with the UDD-6 dataset, achieving a highest effectiveness of 97.0%, followed by ConvUNext at 95.2%, PPO with SAC at 94.3%, TD3 for Dynamic Energy Management at 92.5%, and SAC with MAML for Obstacle Detection at 91.8%.