Scholarlink

This research provides a unique multi-objective optimization (MOO) and reinforcement learning (RL) swarm trajectory planning method for autonomous agents [18]. The approach is based mainly on decentralized communication and adaptive attraction toward the goal following the rule of obstacle avoidance, the agent enables cooperation to optimize trajectory and resources [1]. The study aims to apply this new approach to the Palm harvesting System to optimize critical factors in the field like the dates amount, the timing, and security of the process on agriculture domaine. Instead of delegating the harvesting task to one and only one robot, the idea gives the possibility to collaborate between multiple agents to define the paths, exchange information about already explored area on the field and non-secure trajectories. In addition, adding the learning aspect is a huge plus as it ensures trajectory refining, exploring and real-time adaptation to the new environment. Assuming that in real world the obstacles are not static and palm are not the only objects on the field and harvesting time is crucial in this case.

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

Trajectory Planning Based on RL Swarm Approach Applied to the Palm Harvesting System

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lhoussaine Ait Ben Mouh,
  • Youssef Bouh

摘要

This research provides a unique multi-objective optimization (MOO) and reinforcement learning (RL) swarm trajectory planning method for autonomous agents [18]. The approach is based mainly on decentralized communication and adaptive attraction toward the goal following the rule of obstacle avoidance, the agent enables cooperation to optimize trajectory and resources [1]. The study aims to apply this new approach to the Palm harvesting System to optimize critical factors in the field like the dates amount, the timing, and security of the process on agriculture domaine. Instead of delegating the harvesting task to one and only one robot, the idea gives the possibility to collaborate between multiple agents to define the paths, exchange information about already explored area on the field and non-secure trajectories. In addition, adding the learning aspect is a huge plus as it ensures trajectory refining, exploring and real-time adaptation to the new environment. Assuming that in real world the obstacles are not static and palm are not the only objects on the field and harvesting time is crucial in this case.